Functional Genetic Screening Enables Theranostic Molecular Imaging in Cancer

Targeted therapies developed through mechanism-based drug discovery are revolutionizing cancer care, enabling improved patient outcomes and motivating patient-specific treatment strategies that form the basis for precision medicine (1). Despite this promise, early experiences with these approaches support the need for integrated response assessment to further inform therapy (2). Current theranostic molecular imaging paradigms, including ¹⁸Fluorodeoxyglucose PET, enable the detection of a therapeutic target within the tumor, but fail to inform on the on-target efficacy of the therapeutic. This deficiency has limited the translation of molecular imaging approaches despite the development of a myriad of molecular probes across a spectrum of imaging modalities. Indeed, despite proven survival benefits for targeted therapies, current clinical imaging fails to quantify therapeutic response effectively (3). Integrated theranostic strategies that enable the selection of molecular imaging probes specific for targetable vulnerabilities in cancer hold the potential to overcome this fundamental limitation by providing direct and early quantifications of response.

Massively parallelized biotechnologies are transforming the discovery and development of targeted therapeutics. Among these, genetic screens using CRISPR-based genome editing enable the high-throughput and unbiased identification of potentially targetable vulnerabilities (4). In addition to identifying therapeutic dependencies, existing functional annotations may enable the simultaneous identification of functional imaging probes modified by the drug target. Until recently, existing molecular imaging technologies had limited capacity to assess in vivo enzyme function or inform drug pharmacodynamics given an inability to distinguish the parent substrate from its product. The advent of dynamic nuclear polarization–enhanced ¹³C nuclear magnetic resonance spectroscopic imaging (DNP-MRSI) overcomes this limitation through the ability to measure multiple chemical species simultaneously. DNP is a hyperpolarization technique that increases the fraction of nuclei aligning with an external magnetic field relative to the Boltzmann prediction of thermal equilibrium, producing up to a 10⁵-fold increase in signal-to-noise and enabling the detection of stable ¹³C-labeled isotopes in real-time (5). The role of enzyme abundance and activity in influencing the conversion of ¹³C-labeled substrates suggests that cellular enzymatic dependencies may facilitate the selection of hyperpolarized ¹³C-MRSI probes that specifically measure enzymatic activity (6, 7).

This study hypothesized that CRISPR-based negative selection screening of genes encoding enzymes responsible for hyperpolarized ¹³C-MRS probe metabolism would simultaneously identify therapeutic targets and molecular imaging probes. This strategy holds the potential to enable a novel and high-throughput theranostic paradigm with the power to provide early measures of on-target efficacy and response prediction.

A detailed description of protocols and analyses used for CRISPR screening is provided in the Supplementary Materials and Methods.

All animal experiments were performed according to an institutionally approved protocol (Institutional Animal Care and Use Committee protocol #: 803952) for the safe and humane treatment of animals.

Induction of autochthonous hepatocellular carcinomas in male Wistar rats was performed as described previously with diethylnitrosamine (8). One week prior to completion of the diethylnitrosamine exposure, rats underwent MRI screening approximately twice per week with T₂-weighted MRI [Field of view: 70 mm × 70 mm, grid size: 256 × 256, slice thickness: 2 mm, Repetition time (TR) minimum: 1.4 s (respiratory gated), Echo time (TE): 59.1 ms, and minimum averages: 4) using an Agilent 4.7T 40 cm horizontal bore MR Spectrometer with a 25 gauss/cm gradient tube interfaced to an Agilent DirectDrive console (9). During all scans, induction and maintenance of anesthesia was achieved using approximately 2% isoflurane in O₂, body temperature was maintained at 37°C with a closed loop Heating System (Small Animal Instruments), and respiration rate was monitored. Each animal was positioned within a Polarean proton-tuned (200.1 MHz) birdcage resonator. After individual hepatocellular carcinoma lesions reached approximately 100 mm³ in size, they were used for subsequent study with a lactate dehydrogenase inhibitor (LDHi, 737) graciously provided by Dr. Len Neckers (National Cancer Institute, [email protected]) and the NCI Experimental Therapeutics Program (10, 11). Following inhibitor administration, animals underwent repeated screening with T₂-weighted MRI approximately twice per week.

On the day of therapeutic treatment with a LDHi, each animal was centered within a Polarean proton-tuned (200.1 MHz) birdcage resonator with a ¹³C-tuned (50.525 MHz) surface coil positioned over the right upper quadrant. T₂-weighted imaging was performed to confirm positioning of the surface coil over the tumor of interest using a ¹³C-urea phantom centered within the surface coil (9). Each animal then received a baseline hyperpolarized [1-¹³C]-pyruvate injection followed by a LDHi injection within 15 minutes, and then a follow-up hyperpolarized [1-¹³C]-pyruvate injection within 90 minutes of LDHi administration (n = 9).

The composition and timing of hyperpolarized [1-¹³C]-pyruvate injections were identical prior to and following LDHi administration for each single animal using a concentration of 80 mmol/L hyperpolarized [1-¹³C]-pyruvate. Additional details about injection composition and the hyperpolarization procedure are provided in the Supplementary Materials and Methods. DNP-MRSI was performed using a 2D echo-planar spectroscopic imaging (EPSI) sequence with a spectral bandwidth of 1.1 kHz and 128 points, enabling detection of a chemical shift range of 22 ppm (9, 12). The total field of view was 60 mm × 45 mm with 12 phase-encoded steps and 16 frequency-encoded steps, enabling an in-plane resolution of 3.75 mm × 3.75 mm with a slice thickness matching the thickness of the tumor in the coronal plane (∼6–10 mm). Each image was acquired in approximately 1 second with consecutive images repeated approximately every 2 seconds for a minimum of 20 scans per study.

LDHi injections were performed at a concentration of 10 mg/mL and a dose of either 10 (n = 6) or 20 mg/kg (n = 3). All LDHi injections were performed by hand through tail vein injection at a rate of 1 mL/minute and were followed by a 2 mL saline flush.

Acquired proton images and EPSI spectra were reconstructed in MATLAB 2018B (Mathworks, Inc.). To ensure accurate quantitation of the carbon data, the following corrections were employed: B₁ correction, point spread function correction, polarization efficiency normalization, and baseline subtraction consistent with previously described methods (9). Chemical shifts were determined relative to a ¹³C-urea reference standard (165.5 ppm) present in the ¹³C surface coil. Intratumoral metabolites were determined using region of interest analysis based on registration of EPSI images with T₂-weighted images. To ensure the entire tumor was included in each analysis (including regions encompassing only a partial voxel), the sums of intratumoral resonances were determined from a 2D interpolation of the corrected EPSI data using a cubic convolution (interpolation factor = 8). When the intratumoral signal-to-noise ratio (SNR) for an individual image was ≥2, metabolite quantifications for individual images were calculated as the integral of metabolite peaks identified on the basis of chemical shifts (pyruvate: 170.4 ppm, lactate: 183.35 ppm, and alanine: 176.5 ppm). For individual images within a time series with subthreshold SNR or for which metabolite integrals yielded negative values, metabolite quantifications were recorded as 0 for subsequent analyses. Total ¹³C metabolite (TCM) signal was determined as the sum of quantified pyruvate, lactate, and alanine resonances in an image. Kinetic analyses were performed by taking the area under the curve (AUC) of metabolite quantifications from the dynamic EPSI series prior to ratiometric comparisons. Nonzero AUCs were quantified according to this procedure for all metabolites in all comparisons outlined in this work. Parametric maps were generated from cubic interpolation of metabolite quantifications (interpolation factor = 8).

Statistical analyses were performed in RStudio Desktop 1.1.442; graphical representations of these data were made using GraphPad Prism 8.2.1. Shapiro–Wilk tests for normality were performed for all two-group comparisons. Normally distributed data were compared with Student t tests, while nonnormally distributed data were compared with Wilcoxon signed rank tests. Paired t tests were used for all paired data, while a two-sample t test was used for nonpaired data. T-statistics are reported for all t tests, while V-statistics are reported for all Wilcoxon signed rank tests. Linear regression was used to assess changes in tumor growth from imaging data.

To evaluate the utility of functional genetic screening for therapeutic target identification and probe selection, negative selection CRISPR screens from the DepMap Public 19Q3 dataset, which collectively study the essentiality of 18,333 genes in 625 cancer lines, were analyzed (13, 14). Using the provided cancer type annotations, this analysis isolated differential vulnerabilities, identifying genes encoding proteins most essential to growth (Supplementary Table 1). These data were then filtered to isolate genes encoding differentially essential enzymes for which DNP-MRS probes were previously reported to measure enzymatic activity (Fig. 1A; Supplementary Fig. S1; Supplementary Table S1). Of the 29 different cancer types investigated, 17 demonstrated targetable dependencies that may be imaged using an established ¹³C-labeled DNP-MRS probe (Fig. 1A; Supplementary Fig. S1; Supplementary Tables S1–S3). In addition, analysis of the top five differentially essential genes revealed novel potential imaging targets for four of the remaining cancer types (Fig. 1A; Supplementary Fig. S1; Supplementary Table S1).

As a proof of principle, these analyses were also performed for hepatocellular carcinoma, the most common form of primary liver cancer, which comprise a subset of the liver cancers in the DepMap database. Initial investigation of differentially essential genes identified lactate dehydrogenase A (LDHA), which encodes the predominant isoform of LDH, a known molecular imaging target, as the most differentially essential enzyme-encoding gene (P < 10⁻⁸; Table 1; Fig. 1B; refs. 13, 14). LDHA was the only DNP-MRSI imageable dependency isolated in hepatocellular carcinoma among significantly depleted genes (ref. 15; Supplementary Table S1). The identification of FGFR1, a gene encoding one of several proteins inhibited by the standard-of-care hepatocellular carcinoma therapies lenvatinib and sorafenib, as the third most differentially essential gene further corroborated the screening technique (P < 10⁻⁶; Table 1).

To validate these findings further, a focused CRISPR screening library was designed and adapted to enable high-throughput screening of theranostic probes including guides targeting genes encoding enzymes that metabolize established DNP-labeled molecular imaging substrates, and requiring as few as 4e6 cells (Fig. 1A; refs. 5, 16). Negative selection screens were performed in an established hepatocellular carcinoma cell line included in the DepMap dataset (SNU-449) and the results confirmed the essentiality of LDHA in these cells (log₂ fold-change or LFC = −1.10; P < 10⁻⁴; FDR = 6.30e-5). A negative selection CRISPR screen was then performed in a primary hepatocellular carcinoma cell line (PGM-898) generated in the authors' laboratory from a patient-derived xenograft, further confirming an LDH dependency in hepatocellular carcinoma (LFC = −0.380; P < 0.05; FDR = 0.175).

LDH enzymatically converts pyruvate to lactate, while also regenerating NAD⁺ (Fig. 1C). Hyperpolarized [1-¹³C]-pyruvate is a molecular imaging probe that, when metabolized to form lactate, quantifies local LDH activity (5, 9, 17). To test a theranostic strategy combining LDH inhibition with DNP-MRSI assessment of LDH activity, a translational rat model of hepatocellular carcinoma was utilized wherein autochthonous hepatocellular carcinomas were chemically induced in a background of cirrhosis. To identify whether hyperpolarized [1-¹³C]-pyruvate could be used to measure drug pharmacodynamics, MRSI of hyperpolarized [1-¹³C]-pyruvate was performed before and after intravenous injection of a LDHi at either 10 (n = 6) or 20 mg/kg (n = 3; Fig. 2). As observed in prior studies with this translational model of hepatocellular carcinoma, tumor regions had up to 3-fold greater levels of signal-to-noise than the surrounding liver (9). In all cases, LDHi administration led to the near complete abrogation of intratumoral lactate production, which manifested as reductions in the lactate-to-total carbon metabolite ratio (L/TCM; t = −6.20; df = 8; P < 0.001; n = 9) and the lactate-to-pyruvate ratio (L/P; V = 45; P < 0.01; n = 9; Fig. 2 and 3; Supplementary Fig. S2). Consistent with previous reports of transport-limited metabolism of DNP probes, LDH inhibition also led to increases in the pyruvate-to-total carbon metabolite ratio (P/TCM; t = 4.40; df = 8; P < 0.01; n = 9), the alanine-to-total carbon metabolite ratio (A/TCM; t = 2.82; df = 8; P < 0.05; n = 9), the alanine-to-pyruvate ratio (A/P; t = 2.34; df = 8; P < 0.05; n = 9), and the alanine-to-lactate ratio (A/L; t = 4.35; df = 8; P < 0.01; n = 9; refs. 18, 19).

Given that LDH knockout slowed hepatocellular carcinoma growth in vitro, the therapeutic efficacy of the LDHi was assessed in vivo. Tumor volumes measured from T₂-weighted images acquired before and after inhibitor administration were used to quantify tumor doubling time (20). Tumor volumes from images acquired prior to therapy and those acquired within 10 days following therapy were fitted to exponential growth functions to quantify tumor doubling times. A single dose of the LDHi at either 10 (n = 6) or 20 mg/kg (n = 3) was found to significantly reduce tumor doubling times (V = 44; P < 0.01; n = 9; Fig. 4A and B). The strength of this effect increased when using a 20 mg/kg injection compared with a 10 mg/kg injection (t = 3.87; df = 3.35; P < 0.05; n = 6 at 10 mg/kg and n = 3 at 20 mg/kg; Fig. 4C).

DNP-MRSI data were further analyzed to assess the ability to predict therapeutic response at the time of drug administration. Interestingly, intratumoral alanine production after LDHi administration robustly predicted the degree of growth inhibition (P < 0.001). In addition, the intratumoral level of alanine following LDHi administration was confirmed as a predictive biomarker when controlling for multiple potential covariates, including models correcting for postinhibitor pyruvate levels (P < 0.01; Table 2), preinhibitor alanine levels (P < 0.01), and postinhibitor pyruvate and lactate levels (P < 0.01). These data demonstrate that the postinhibitor alanine level directly correlates with the observed LDHi-induced reduction in tumor growth. Postinhibition intratumoral lactate, which uniformly approached the limit of detection, did not predict response, likely secondary to insufficient sensitivity.

Taken together, these data demonstrate that functional genetic screening informs a theranostic paradigm leveraging DNP-MRSI to assess drug pharmacodynamics, as well as to predict therapeutic efficacy.

Functional genetic screening identified imageable metabolic vulnerabilities in 17 unique cancer types. Among these, LDH was identified as a differential vulnerability in hepatocellular carcinoma. These data demonstrate that LDH is more critical to hepatocellular carcinoma growth than it is to the growth of other cancers (21). LDHA was found to be the second most differentially essential protein-encoding gene and the single most differentially essential enzyme-encoding gene for hepatocellular carcinoma cell growth. The dependence of hepatocellular carcinoma on LDH was further validated using a focused CRISPR screen in two cell lines, including one derived from a patient biopsy. These findings underscore the growing recognition that cancer cells demonstrate differential dependencies on metabolic enzymes (14, 22, 23). Consistent with the hypothesis that differential enzyme essentiality enables a theranostic imaging approach via simultaneous identification of an imaging probe and a therapy, genetic screening accurately identified hyperpolarized [1-¹³C]-pyruvate and its metabolites as in vivo biomarkers of on-target efficacy of, and response to, LDH inhibition in hepatocellular carcinoma.

While the role of LDH in promoting cancer growth is well established, these findings underscore the importance of variabilities in enzymatic dependencies and activities both within and across cancer types. This variability forms the basis for clinical decision-making both with respect to the selection of targeted cancer therapies, as well as the selection of imaging studies (24, 25). Indeed, variability in the expression and activity of the GLUT1 transporter and hexokinase underlie differences in ¹⁸FDG-avidity on PET imaging and influence heterogeneity in the application of ¹⁸FDG-PET in cancer imaging (25–28). The presented data suggest that genetic screening tools can be applied to characterize these dependencies and enable the selection of optimal imaging probes on a cancer-by-cancer basis. Given inherent variability in cancer dependencies, curated screens, including those performed with the DNP-focused library, minimize experimental complexity and requisite cell number, enabling confirmation of potential therapeutic and imaging targets in patient-derived samples. Moreover, unbiased screening, including screens available within the DepMap datasets, can independently isolate alternative dependencies, suggesting avenues for future therapy and imaging probe development. Indeed, a wider, unbiased analysis limited to the top five differentially essential genes, identified potential metabolic imaging targets in four cancer types for which the focused DepMap screens did not identify a dependency on an enzyme that metabolizes an existing DNP-probe. While this study leveraged the unique capability of DNP-MRS imaging to determine enzyme function based on direct measurements of the administered substrate and its metabolites, the described paradigm is generalizable to other imaging modalities. This screening approach can be readily expanded to include targets of other molecular imaging probes including the extensive repertoire of PET-based radiotracers already developed for oncologic imaging (29).

In addition, by directly imaging the metabolic pathway targeted by a therapy, the presented data demonstrate a role for metabolic molecular imaging for predicting a clinical phenotype at the time of therapy administration. In this study, alanine generation after LDH inhibition predicted the reduction in tumor doubling time. This finding, in combination with the observed dose-dependent growth inhibition, underscores the importance of direct assessments of on-target efficacy for metabolic inhibitors. Indeed, suboptimal on-target efficacy has limited the translation of LDH inhibitors described in the literature to date (11, 30). As such, these data, in combination with previous studies, suggest the unique potential of metabolic imaging with DNP-MRSI of hyperpolarized [1-¹³C]-pyruvate to inform targeted therapy selection, therapeutic efficacy, as well as disease severity on a broader scale (23, 31). As a highly regulated metabolic enzyme, assessments of LDH activity in vivo with hyperpolarized [1-¹³C]-pyruvate MRSI may stratify patients and predict response to targeted systemic or locoregional therapies beyond LDH inhibitiors (9, 32). In hepatocellular carcinoma, hyperpolarized [1-¹³C]-pyruvate MRSI may predict response to lenvatinib and sorafenib given prior literature demonstrating the prognostic value of serum LDH concentration (33, 34). Similarly, a recent clinical trial in prostate cancer demonstrated that lactate generation from DNP-[1-¹³C]-pyruvate predicts tumor grade (35). Future preclinical and clinical studies will help assess the predictive value of hyperpolarized [1-¹³C]-pyruvate MRSI in this broader context. Moreover, the presented data emphasize a role for DNP-MRSI in this capacity beyond hyperpolarized [1-¹³C]-pyruvate. The identified dependence of hepatocellular carcinoma on FGFR1 suggests that imaging FGFR1-encoded protein activity could provide a direct assessment of lenvatinib and sorafenib efficacy in hepatocellular carcinoma. While there is not yet a DNP-labeled probe for this application, ligand-receptor–induced changes in T₁ or chemical shift may guide the development of DNP-MRS probes that report on the efficacy of receptor targeting (36). Furthermore, as MRSI can be performed with multiple DNP-labeled probes simultaneously, optimization of probe combinations may facilitate diagnosis based on independent cancer dependencies. For example, the included analysis of the DepMap database suggest that DNP-[2-¹³C]-dihydroxyacetone, together with hyperpolarized [1-¹³C]-pyruvate, may enable the noninvasive distinction of colorectal cancer, which frequently metastasizes to the liver, from hepatocellular carcinoma (Fig. 1; Supplementary Fig. S1; Supplementary Table S1).

Importantly, the described data hold important implications for the treatment of hepatocellular carcinoma, the most rapidly increasing cause of cancer-related mortality in the United States (37). The majority of patients with hepatocellular carcinoma present with incurable disease at diagnosis and, despite the approval of targeted therapies, life expectancy remains less than 20 months (38). This dismal prognosis issues from challenges in developing targeted therapies and a lack of functional imaging strategies that report on therapeutic efficacy (3, 39–42). This study leveraged DNP-MRSI to visualize intratumoral metabolism of hyperpolarized [1-¹³C]-pyruvate, confirming that LDH inhibition prevented lactate production and reduced tumor doubling times in vivo. A single dose of a LDHi in this model of hepatocellular carcinoma slowed tumor growth in a dose-dependent manner suggesting that higher doses, repeat injections, or local administration may enable tumor regression and/or mitigate hepatocellular carcinoma recurrence. Previous studies have demonstrated that LDH inhibition may reduce tumor doubling times in several cancers and that myc-driven tumors, including hepatoblastoma, have enhanced LDH activity; however, in vivo LDH targeting in hepatocellular carcinoma has not been demonstrated previously (5, 43, 44). In addition, while the observed LDHi-induced reduction in lactate production was anticipated, the degree of LDH inhibition observed in our study is unique (23, 43, 45). These data, in combination with the genetic screening data, demonstrate that hepatocellular carcinoma is more susceptible to LDH inhibition than other cancers. Consistent with this hypothesis, a recent study of pancreatic cancer found that larger and more frequent doses of the LDHi described herein were required to observe growth inhibition of xenografts (43). This discrepancy suggests that pancreatic cancer is less dependent upon LDH for growth than hepatocellular carcinoma (Fig. 1B). Future work will determine the optimal dose, frequency, and delivery method for a LDHi by leveraging the described DNP-MRSI analysis pipeline.

In summary, this study demonstrates that biologically motivated therapeutic and imaging probe selection through genetic screening enables a high-throughput theranostic paradigm in oncology. In the era of precision medicine, wherein patient-derived samples are increasingly utilized to inform management, this approach represents a novel strategy to overcome existing barriers and facilitate the application of molecular imaging at the bench as well as at the bedside. Finally, the described data provide a blueprint for validation of this approach through the integration of hyperpolarized [1-¹³C]-pyruvate MRSI in clinical trials of LDH inhibitors.

N.R. Perkons reports grants from NIH NCI (F30CA232388) and grants from Society of Interventional Radiology Foundation (Allied Scientist Training Grant) during the conduct of the study, as well as non-financial support from Guerbet LLC (Travel to a research meeting) outside the submitted work. D. Ackerman reports other from Guerbet (Conference attendance) outside the submitted work. T.P.F. Gade reports grants from NIH (DP5 OD021391) during the conduct of the study, as well as grants from Society for Interventional Radiology, grants from Society for Interventional Oncology, grants from Radiological Society of North America, grants from NIH, and personal fees from Trisalus Life Sciences (Scientific Advisory Board) outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

N.R. Perkons: Conceptualization, resources, data curation, software, formal analysis, funding acquisition, validation, investigation, methodology, writing-original draft, writing-review and editing. O. Johnson: Data curation, investigation, methodology, writing-review and editing. G. Pilla: Data curation, investigation, methodology, writing-review and editing. E. Profka: Data curation, investigation, methodology, writing-review and editing. M. Mercadante: Data curation, investigation, methodology, writing-review and editing. D. Ackerman: Conceptualization, supervision, funding acquisition, project administration, writing-review and editing. T.P.F. Gade: Conceptualization, resources, formal analysis, supervision, funding acquisition, methodology, writing-original draft, project administration, writing-review and editing.

This study was supported in part by the NIH, NCI, Grant F30CA232388 and Society of Interventional Radiology Foundation's Allied Scientist Grant (all to N.R. Perkons). This study was also supported in part by the NIH, Director's Office, Grant DP5OD021391 (to T.P.F. Gade).

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.