Photoacoustic imaging of elevated glutathione in models of lung cancer for companion diagnostic applications

Abstract

Companion diagnostics (CDx) are powerful tests that can provide physicians with crucial biomarker information that can improve treatment outcomes by matching therapies to patients. Here, we report a photoacoustic imaging-based CDx (PACDx) for the selective detection of elevated glutathione (GSH) in a lung cancer model. GSH is abundant in most cells, so we adopted a physical organic chemistry approach to precisely tune the reactivity to distinguish between normal and pathological states. To evaluate the efficacy of PACDx in vivo, we designed a blind study where photoacoustic imaging was used to identify mice bearing lung xenografts. We also employed PACDx in orthotopic lung cancer and liver metastasis models to image GSH. In addition, we designed a matching prodrug, PARx, that uses the same S_NAr chemistry to release a chemotherapeutic with an integrated PA readout. Studies demonstrate that PARx can inhibit tumour growth without off-target toxicity in a lung cancer xenograft model.

Despite progress in the diagnosis, treatment and management of cancer, nearly 10 million individuals will die each year, with lung cancer being the leading cause¹. An emerging technology that can potentially improve these statistics is companion diagnostics (CDx)²—tests that provide essential information for determining which patients may benefit from a particular treatment by detecting biomarkers that are critical for drug activation³. Current US Food and Drug Administration (FDA)-approved CDx are designed for in vitro testing and thus cannot account for variations in the tumour microenvironment that can impact drug efficacy. To provide accurate, real-time information, it is crucial to develop new molecular imaging-based CDx, as well as corresponding drugs for aberrant cancer properties. However, most clinically approved imaging agents for cancer target cell surface biomarkers through stoichiometric binding events. In addition to poor signal-to-noise, binding-based strategies preclude the detection of small-molecule targets, such as glutathione (GSH). Activity-based sensing (ABS) is a powerful alternative that relies on the selective detection of chemical reactivity. In contrast to fluorescence-based ABS probes, which can only operate in the millimetre depth range due to scattering and light attenuation, the corresponding chemical tools for photoacoustic (PA) imaging can be employed as a CDx for in vivo testing. PA is a non-invasive modality that commonly utilizes near-infrared (NIR) light to induce the generation of an ultrasound signal. Because clinically relevant ultrasound frequencies travel through biological tissues with minimal perturbation, it is possible to pinpoint the source of the signal to achieve high-resolution images (tens of micrometres) in the centimetre range⁴. Label-free PA imaging has already been employed clinically for the detection of breast cancer⁵, thyroid cancer⁶, inflammatory arthritis⁷ and scleroderma⁸. Recently, our group and others have expanded the scope of PA imaging through the development of acoustogenic ABS probes^9,10. Select examples include imaging agents for the detection of metal ion dysregulation^11–15, hypoxia^16–20, proteases^21,22 and signalling molecules (for example, nitric oxide^23–25 and hydrogen sulfide²⁶). In this Article, we leverage classic principles in physical organic chemistry to guide the development of the first PA imaging-based CDx (PACDx), as well as a matching gemcitabine-based prodrug (PARx). Application of PACDx and PARx in an unprecedented blind study allowed us to successfully identify and treat mice with lung cancer, respectively. Additionally, our new CDx was employed in orthotopic lung cancer and liver metastasis models to demonstrate its broad utility. This application is important, because it overcomes an outstanding challenge in the field of PA imaging, which is to detect biomarkers in the lung region.

Results

Physical organic approach to guide the design of PACDx.

GSH is an abundant biological thiol that is essential for maintaining redox homeostasis and detoxifying xenobiotics²⁷. Aberrant changes in the cellular levels of GSH are correlated with a variety of pathologies, such as cancer²⁸. Among the various cancer types, non-small cell lung carcinoma has the largest change in GSH levels in both patient-derived tissue samples (up to fourfold)²⁹ and human cell lines (up to sevenfold)^30,31, relative to healthy tissue and non-cancerous cell lines, respectively. In this study, we chose to develop a CDx assay for lung cancer based on the detection of GSH via PA imaging. Importantly, even at the onset of cancer, GSH levels are elevated, presenting exciting opportunities for early cancer detection³².

The design of PACDx features two key elements, a PA dye capable of generating a strong PA signal upon irradiation and a trigger for detecting GSH in its reduced form (Fig. 1a). In addition to its large extinction coefficient (ε = >10⁴ M⁻¹ cm⁻¹) in the near-infrared region, the hemicyanine dye (HD) platform was selected because of its lipophilic cationic character, which can enhance uptake by tumour cells and render the resultant probe a poor substrate for glutathione S-transferase (GST), an abundant cytosolic enzyme that catalyses the conjugation of GSH to electrophilic centres^33,34. We deemed this to be important, because the expression of GST may vary in cancer and, therefore, cross-reactivity can confound the imaging results. We employed a physical organic approach to develop a precision-tuned, GSH-responsive trigger that can distinguish GSH levels between lesions and healthy tissue. This represents a substantial challenge, because existing triggers based on disulfide exchange³⁵ and S_NAr³⁶ chemistry are too reactive. For example, installation of the common 2,4-dinitrobenzenesulfonate trigger onto our PA platform afforded a probe that could not distinguish GSH levels across several mammalian cell lines (Supplementary Fig. 1), because it was fully activated by only 100 μM GSH within several minutes (~10-fold less than physiological levels; Fig. 1f). To establish a structure–reactivity relationship, we synthesized a panel of nine probes featuring a range of electron-withdrawing and -donating groups at the para position. The resultant Hammett plot revealed a linear correlation between log(k_X/k_H) and the corresponding σ⁻ constants (k_X and k_H are rate contants for when the para-substituent is X or H, respectively; σ is the substituent constant) (Fig. 1b)³⁷. A positive ρ value of 1.10 indicates that electron-deficient substituents at this position were more efficient at stabilizing the Meisenheimer complex, resulting in enhanced reactivity (ρ is the reaction constant). Unfortunately, the most reactive probe in this series (p-nitro) was only partially activated, to give a 5.6-fold turn-on response after treatment with 10 mM GSH for 1 h (Fig. 1g). However, a dose- and time-dependent response was evident. Encouraged by these results, we synthesized an additional four probes with both ortho- and para-electron-withdrawing groups to further increase the S_NAr reactivity (Supplementary Fig. 2). As anticipated from our structure–reactivity data, installation of a fluoro substituent at the ortho position resulted in a highly optimized trigger (Fig. 1h). Activation of the resulting probe, PACDx, under physiological conditions was completely attenuated. On the other hand, a dose-dependent signal increase of up to 31.6-fold was observed when PACDx was incubated with 10 mM GSH for 1 h. Rate constants for the S_NAr reaction for PACDx, the p-nitro analogue and the 2,4-dinitro congener were measured at various temperatures to construct the corresponding Eyring plots (Fig. 1c). The derived activation parameters for PACDx are ΔH^‡ = 70.5 kJ mol⁻¹ and ΔS^‡ = −117.1 J mol⁻¹ K⁻¹. These data suggest that the S_NAr reaction involves a highly ordered transition state, which probably results from effective stabilization of the Meisenheimer complex. The Gibbs free energy of activation (ΔG^‡) for PACDx was calculated to be 106.8 kJ mol⁻¹, which indicates that the S_NAr reaction was energetically more favourable than the p-nitro analogue, but less favourable compared to the 2,4-dinitro congener. Finally, we constructed two Brønsted relationship graphs where the pK_a values of the corresponding phenols were plotted against ΔG^‡ (Fig. 1d) and log(k) (Fig. 1e). In both cases, we observed a linear correlation with R² values of 1.00, suggesting that this information can guide probe design involving S_NAr chemistry. All pertinent results are summarized in Supplementary Table 1.

Fig. 1 |. Application of physical organic chemistry to tune S_NAr reactivity for GSH sensing.

a, General schematic for GSH-responsive photoacoustic companion diagnostics. b, Hammett plot for the S_NAr reaction between para-substituted benzenesulfonyls and 10 mM GSH at 120 °C. The dotted line is the best linear fit of the rate constant data. k_X and k_H are rate constants for when the para-substituent is X or H, respectively. c, Eyring plots for 2, 5 and PACDx reacting with GSH (structures of 2, 5 and PACDx are shown as insets in f–h). The dotted lines represent the best linear fits. k, rate constant; T, temperature; k_B, Boltzmann constant; h, Planck’s constant. d,e, Brønsted plots indicating the linear relationship between ΔG (d) and log(k) (e) and pK_a (R² = 1.0 and 0.99, respectively) for 2, 5 and PACDx reacting with GSH at 37 °C. f–h, Dose-dependent activation of 2 (f), 5 (g) and PACDx, with [GSH] of 0.1–10 mM (h). A and A₀ are defined as the final and initial absorbance values, respectively. All assays were performed at pH 7.4, 70% PBS/MeCN. Data are presented as mean values ± s.d. (n = 3 independent experiments).

Evaluation of PACDx in vitro and in live cells.

Before activation, PACDx does not absorb strongly within the PA window (680–950 nm). However, treatment with GSH induces a bathochromic shift (Fig. 2a). Irradiation at λ_max (690 nm) yields the strongest PA response in vitro (Fig. 2b). A dose-dependent signal enhancement with a limit of detection of 0.39 mM was observed when PACDx was treated with GSH (Fig. 2c,d). By contrast, there was no notable reactivity when the probe was incubated with cysteine (200 μM) or homocysteine (100 μM) (Supplementary Fig. 5). Next, we evaluated the performance of PACDx in A549 lung cancer cells, U87 glioblastoma cells and HEK 293 cells (a non-cancerous cell line) using confocal microscopy. PACDx was highly fluorescent in A549 cells, with a cytosolic and mitochondrial staining pattern (Fig. 2e). However, when cells were pretreated with N-ethylmaleimide, a reagent used to reduce the levels of intracellular thiols, the fluorescence signal was attenuated by ~50% (Fig. 2f). To confirm that the decrease in intensity was due to GSH depletion, a third set of cells were treated with a non-responsive control probe (Ctrl-PACDx, Supplementary Fig. 6 and Fig. 2g,h). A trend comparable to that for A549 cell imaging was observed in U87 cells (Supplementary Fig. 7). When A549 and U87 cells were pretreated with ethacrynic acid³⁸, a potent reversible inhibitor of human GST, no effect on probe activation was observed, which suggests that activation of PACDx is independent of GST enzymatic activity (Supplementary Fig. 8).

Fig. 2 |. Evaluation of PACDx GSH-responsiveness in vitro and in in cellulo systems.

a, Absorbance profile of 5 μM PACDx before (black line) and after (red line) treatment with 10 mM GSH. b, PA spectra of 50 μM turned-over HD dye. c,d, PA signal (n = 3 independent experiments) (c) and PA images of PACDx (d) in response to GSH. Samples were irradiated at 690 nm. All in vitro assays were performed at 37 °C, pH 7.4, with 70% PBS/MeCN, and data were collected after 1 h. e–g, Confocal microscopy image representing A549 cells treated with 5 μM PACDx for 1 h at 37 °C (e), A549 cells pretreated with 1 mM N-ethylmaleimide (NEM) then incubated with 5 μM PACDx for 1 h at 37 °C (f) and A549 cells treated with 5 μM Ctrl-PACDx for 1 h at 37 °C (g). Scale bar, 20 μm. h, Normalized fluorescence intensity obtained from cell imaging under the conditions in e–g. Data are presented as mean values ± s.d. (n = 12 biologically independent samples). i, Normalized fluorescence intensity obtained from cell imaging A549, U87 and HEK 293 cells with 5 μM PACDx for 1 h at 37 °C. Data are presented as mean values ± s.d. (n = 12 biologically independent samples). j, Normalized [GSH] obtained from Ellman’s assay. Data are presented as mean values ± s.d. (n = 2 biologically independent samples). Statistical analysis was performed using a two-tailed Student’s t-test (α = 0.05), ***P < 0.001, **P < 0.01, *P < 0.05. NS, not significant.

For PACDx to function as an effective CDx, it is essential for it to be able to accurately differentiate the relative GSH levels in lung cancer cells compared to healthy cells. The intensity of PACDx in A549 cells was indeed the highest relative to U87 and HEK 293 cells (Fig. 2i). However, it is important to note that the signal from the U87 cells was notably lower than that of the HEK 293 cells. To account for possible variations in probe uptake and retention, we incubated PACDx with cell lysates and obtained consistent results (Supplementary Fig. 9). This suggests that the intracellular levels of GSH are highest in A549 cells and lowest in U87 cells. This was confirmed with the established Ellman’s assay, which allows for accurate quantification of GSH levels in cell lysates (Fig. 2j)³⁹.

Design and synthesis of PARx.

Next, we turned our attention to designing a prodrug invoking the same GSH-mediated chemistry. We rationalized it would be possible to append a chemotherapeutic to the PACDx core by strategically installing a hydroxymethyl handle ortho to the phenolic alcohol. On removal of the trigger, the resultant phenolate intermediate can fragment via a 1,4-elimination pathway to release the drug and the corresponding dye (HD-CH₂OH) for PA imaging (Fig. 3a). We chose to append gemcitabine, an FDA-approved drug, because it is commonly used to treat non-small cell lung cancer through the inhibition of DNA synthesis^40,41. However, as with many chemotherapeutics, gemcitabine indiscriminately targets any rapidly dividing cell in the body, which results in adverse effects such as severe liver toxicity⁴². Masking the primary alcohol with a variety of capping groups has led to the development of gemcitabine-based prodrugs displaying attenuated cytotoxicity until they are activated⁴³. Although we could have directly modified gemcitabine with our new 2-fluoro-4-nitrobenzenesulfonyl trigger, we wanted to utilize the lipophilic cationic dye for targeting and to leverage the PA imaging capabilities of the resulting prodrug (herein named PARx) to monitor drug release in real time.

Fig. 3 |. Development of PARx, a gemcitabine-based prodrug with matching GSH reactivity.

a, Reaction scheme of PARx with GSH to release HD-CH₂OH and gemcitabine. b, Normalized absorbance profile of 5 μM PARx before (black line) and after (red line) treatment with 10 mM GSH at 37 °C for 1 h (pH 7.4, 70% PBS/MeCN). c, PA spectra of 50 μM turned-over HD-CH₂OH dye in 70% PBS/MeCN. d, PA images of PARx in response to GSH. Samples were irradiated at 690 nm. e, Reactivity of PARx with biologically relevant metals, amino acids, rat liver microsomes (RLM), ascorbic acid (AA), reactive nitrogen species, reactive oxygen species and thiols after incubation for 1 h at 37 °C. f, Normalized fluorescence intensity obtained from cell imaging A549, U87 and HEK 293 cells with 5 μM PARx for 1 h at 37 °C. Data are presented as mean values ± s.d. (n = 12 biologically independent samples). Statistical analysis was performed using a two-tailed Student’s t-test (α = 0.05), ***P < 0.001, **P < 0.01, *P < 0.05. g, Cell viability assay using various concentrations of gemcitabine, PARx and Ctrl-PARx-2 after a 48-h incubation with A549 cells. Data are presented as mean values ± s.d. (n = 6 biologically independent samples). Statistical analysis was performed using a two-way analysis of variance with a Dunnett’s multiple comparisons test (α = 0.05); ****adjusted P > 0.0001.

PARx was synthesized starting from the sequential reduction of 2,4-dihydroxybenzaldehyde and tert-butyl(dimethyl)silyl protection to afford 16 in 61% yield (Supplementary Information). In situ deprotection of the phenolic alcohols with sodium hydride facilitated a nucleophilic substitution and retro-Knoevenagel sequence with Cy7-Cl to obtain 17. The GSH-responsive trigger was installed, and the primary alcohol was deprotected under acidic conditions to yield 18 (Ctrl-PARx-2) in 46% over two steps. Finally, a chloroformate intermediate was generated using phosgene and then reacted with gemcitabine to obtain PARx (Supplementary Information).

Evaluation of PARx in vitro and in live cells.

With PARx in hand, we first confirmed that attaching gemcitabine did not alter the PA imaging properties or the responsiveness of the GSH trigger (Fig. 3b–d and Supplementary Fig. 10). Interestingly, we observed that PARx consistently yielded a more intense PA signal compared to PACDx, presumably due to the larger PA brightness value (ɛ × (1 − Φ_Fl))¹⁹, where Φ_Fl is fluorescence quantum yield. We also demonstrated that PARx displayed exceptional selectivity against a panel of metal ions, amino acids, reductants, reactive nitrogen and oxygen species, metabolic liver enzymes and competing thiols (Fig. 3e). Next, we employed MS and NMR analyses to confirm that GSH mediates the release of gemcitabine (Supplementary Figs. 11 and 12). To further support these results, we prepared a control compound (Ctrl-PARx-1) that is equipped with an unreactive 4-methoxybenzenesulfonyl trigger (Supplementary Fig. 13). MS analysis revealed that both the trigger and carbonate linkage of Ctrl-PARx-1 were stable in the presence of 10 mM GSH for at least 1 h (Supplementary Fig. 14). Moreover, PARx could readily distinguish GSH levels in live A549, U87 and HEK 293 cells, independent of GST activity (Fig. 3f and Supplementary Figs. 16 and 17). Finally, we assessed the cytotoxicity of PARx in A549 cells using the MTT assay and observed dose-dependent toxicity that is comparable to free gemcitabine. By contrast, with treatment with either Ctrl-PARx-1 or Ctrl-PARx-2, the cytotoxicity was significantly attenuated (Fig. 3g). MTT assays in U87 cells established that PARx requires the elevated levels of GSH found in A549 cells to effectively mediate gemcitabine release (Supplementary Fig. 18).

In vivo imaging of GSH and tumour inhibition.

To evaluate the in vivo efficacy of the PACDx and PARx pair, we established an A549 xenograft model of lung cancer. After the tumours had grown to ~100 mm³, we administered PACDx via intratumoral injection. Irradiation at 680 nm (the optimal in vivo wavelength) 1 h post-injection resulted in a ~1.5-fold PA turn-on response relative to the control flank. Next, we introduced PARx via systemic administration to determine its biodistribution profile, as well as potential off-target cytotoxic effects. We collected tissue from vital organs 1 h post-injection for PA imaging analysis. The average PA signals from the heart, kidneys, liver and spleen were comparable between the treatment and control groups (Fig. 4a,b). By contrast, the average PA intensity was 1.2-fold higher in the tumour for the PARx-treated animals. This trend suggests selective activation and drug release. Moreover, these results also suggest PARx will not have off-target toxicity. This was confirmed by performing histological staining, which revealed that there was minimal toxicity in all organs examined (Fig. 4c). However, PARx was highly toxic in tumour tissue, as indicated by the decrease in the number of stained nuclei. In addition, terminal deoxynucleotidyl transferase dUTP nick end labelling (TUNEL) staining revealed that PARx-treated tumours experienced significantly greater cell death via apoptosis (54.3 ± 7.4%) compared to control tumours (1.9 ± 1.9%) (Fig. 4d,e).

Fig. 4 |. Assessment of PARx biodistribution and in vivo activation.

a, Representative ex vivo PA images of heart, kidneys, liver, spleen and tumour after systemic injection of PARx (400 μM, 10% dimethyl sulfoxide/phosphate buffered saline (DMSO/PBS), retro-orbital injection) or vehicle. Samples were irradiated at 680 nm. Scale bar, 2 mm. b, Normalized PA signal relative to the PA signal at 900 nm (PARx does not absorb at this wavelength) after systemic injection of PARx or vehicle. Data are presented as mean values ± s.d. (n = 3 independent animals). c, Haematoxylin and eosin (H&E) staining of heart, kidney, liver, spleen and tumour tissue from PARx-treated and untreated A549 xenografts. Scale bar, 100 μm. d, TUNEL staining of tumour tissue from PARx-treated and untreated A549 xenografts. Brown staining indicates apoptotic cell death. Scale bar, 25 μm. e, Quantification of TUNEL staining. Data are presented as mean values ± s.d. (n = 6 biologically independent samples). Statistical analysis was performed using two-tailed Student’s t-test (α = 0.05), ***P < 0.001.

Over a 21-day period, PARx was intratumorally injected into A549 xenografts every seven days. For this initial experiment, we opted for local administration because we wanted to bypass systemic and metabolic processes. PA imaging was performed 1 h after treatment to monitor gemcitabine release (Supplementary Fig. 19). Compared to control tumours (Fig. 5a,d), the PA signal was higher in PARx-treated tumours. This indicates PARx activation (Fig. 5b,d), leading to complete attenuation of tumour growth (Fig. 5e). By contrast, tumours grew up to ~600 mm³ in size when animals received a vehicle control. To evaluate the systemic compatibility of PARx, we retro-orbitally administered PARx every seven days over the course of a 21-day period (Fig. 5e). Of note, PA monitoring was not employed during the treatment period beyond the first imaging session. The inhibition of tumour growth was nearly identical to the results obtained via intratumoral administration. The body weight of each animal was monitored as a measure of general toxicity. Under no treatment conditions did we observe any loss of weight (Fig. 5f) or change in behaviour⁴⁴. To further challenge the selectivity of PARx, we increased the dosing frequency to once every three days (Supplementary Fig. 20). We hypothesized that we would not see any adverse effects, especially in the liver, where severe damage is common with free gemcitabine. Indeed, we did not observe a change in the body weight (Supplementary Fig. 21) or liver damage.

Fig. 5 |. Determination of PARx efficacy in a murine model of lung cancer.

a–c, PA images of tumours after treatment with vehicle (10% DMSO/PBS) (a), intratumoral injection of PARx (100 μM, 10% DMSO/PBS) (b) or retro-orbital injection of PARx (400 μM, 10% DMSO/PBS) (c). Samples were irradiated at 680 nm. Scale bar, 2 mm. d, PA signal after treatment with vehicle (n = 6 independent animals) or PARx (^aintratumoral injection, n = 3 independent animals; ^bretro-orbital injection, n = 6 independent animals). Data are presented as mean values ± s.d. (imaging performed when the tumour volume was <200 mm³). e, Average tumour volume after treatment with vehicle (n = 6 independent animals) or intratumoral (n = 3 independent animals) or retro-orbital injection (n = 6 independent animals) of PARx over 21 days. f, Average body weight after treatment with vehicle or intratumoral or retro-orbital injection of PARx over 21 days. g, Representative tumours that were treated with vehicle (n = 6 independent animals), intratumoral injection of PARx (n = 3 independent animals) or retro-orbital injection of PARx (n = 6 independent animals). Statistical analysis was performed using the Kruskall–Wallis test (Bonferroni correction: α = α/n = 0.05/3 = 0.01667).

Application of PACDx in a blind study.

To evaluate the performance of PACDx as a CDx, we designed a blind study to identify animals with lung cancer based on GSH levels. First, a group of nude mice were implanted with either A549 or U87 cells. After the animals were tagged and randomized by a second researcher, their identity was concealed until completion of the study. The purpose of implanting U87 cells, as opposed to no tumours, was to ensure that there would be no biasing of the results based on appearance. Next, we measured the PA intensity change after administration of a vehicle to healthy tissue (1.06 ± 0.17, n = 9) to establish a diagnostic threshold (mean value + 2 s.d.). After the tumours had grown to ~100 mm³, PACDx was administered to perform PA imaging (Fig. 6a). We identified one animal that was one s.d. above the mean (1.30) and two animals two s.d. above the mean (1.47 and 1.61) (Fig. 6b). These three subjects (out of seven, total) were categorized to ‘group 1’, which we assigned to be mice bearing A549 tumours. The remaining four animals (0.84, 1.05, 1.06 and 1.07) were grouped into ‘group 2’ (Fig. 6b). Although the aim of the study was to demonstrate that PACDx can be used to identify animals that would respond to PARx, we decided to treat both groups with the prodrug. We hypothesized that prodrug treatment of the first group would result in substantial tumour attenuation. Moreover, we predicted that the administration of PARx to the second group would afford no PA signal enhancement due to lower intratumoral GSH levels. All animals received PARx via retro-orbital injection once every three days for 21 days. Representative images of tumours from group 1 and 2 are shown in Fig. 6c. At the completion of the study, no notable increase in tumour size was observed in group 1, whereas the tumour volume of group 2 had increased to over 1,000 mm³ (Fig. 6d) and the mice had lost substantial body weight (Fig. 6e). It is possible that gemcitabine is intrinsically less effective for U87 tumours. In addition, we cannot rule out a scenario where uptake of PARx in group 1 tumours is greater. However, we can conclude that PARx is not being activated in the U87 tumours owing to a lack of PA signal enhancement in vivo, which is supported by the in vitro and cell-based assays we have previously performed. When the identity of the animals was revealed, we found that we were accurate in each instance, with six of seven mice correctly assigned with greater than 95% confidence.

Fig. 6 |. Application of PACDx and PARx in unbiased animal studies.

a, Schematic illustrating the stratification of a patient group based on PA signal intensity in a blind study. b, Stratification based on PA fold turn-on after retro-orbital injection of PACDx (400 μM, 150 μl, 10% DMSO in saline). The diagnostic threshold was defined as the mean PA fold turn-on after retro-orbital injection of vehicle (150 μl, 10% DMSO in saline, n = 9 independent animals) in healthy mice +2 s.d. to obtain 95% confidence. The black dashed line indicates the diagnostic threshold and the grey dashed lines indicate s.d. Normalized PA turn-on of stratified groups 1 (n = 3 independent animals) and 2 (n = 4 independent animals) are shown. Data are presented as mean values ± s.d. c, Representative PA images of stratified groups 1 (n = 3 independent animals) and 2 (n = 4 independent animals) after administration of PACDx. Samples were irradiated at 680 nm. Scale bar, 2 mm. d,e, Average tumour volume (d) and body weight (e) of stratified groups 1 (n = 3 independent animals) and 2 (n = 4 independent animals) during a 21-day treatment period with PARx. Data are presented as mean values ± s.d. f, Cross-section schematic of a mouse to reference the positioning of the lungs and tumour. g, Representative spectrally unmixed multi-spectral optoacoustic tomography (MSOT) images of the lungs 1 h post-injection of vehicle (150 μl, 10% DMSO in saline) or PACDx (400 μM, 150 μl, 10% DMSO in saline). Scale bar, 5 mm. h, Average PA signal from MSOT imaging (n = 5 independent animals). Statistical analysis was performed using a two-tailed Mann–Whitney U test (α = 0.05, P = 0.007).

Evaluation of PACDx in orthotopic lung cancer and liver metastasis models.

PA imaging of the lung is an outstanding challenge due to a limited acoustic window through the rib cage and attenuation of the acoustic signal by gas present in air sacs. However, to demonstrate the potential versatility of PACDx, it is essential for us to evaluate its performance in relevant lung cancer models. To this end, we utilized A549 luciferase cells to generate orthotopic lung tumours for PA imaging. Luciferase-expressing cells were used because bioluminescence can be employed to visualize whether the procedure was successful and to subsequently track tumour growth. Before inoculation, we determined that the levels of GSH in wild-type and A549 luciferase cells were indistinguishable (Supplementary Fig. 27). Six weeks after inoculation, IVIS imaging showed a strong bioluminescent signal in the lung region, indicating successful inoculation (Supplementary Fig. 28). Cross-sectional PA imaging was then performed (Fig. 6f) to reveal a vascularized region where the cancer cells were inoculated (Supplementary Fig. 30). The identity of the tumours was confirmed ex vivo at the conclusion of the study (Supplementary Fig. 31). After initial background scans were acquired, PACDx (or a vehicle control) was administered systemically and final images were recorded after 1 h. All data were processed via spectral unmixing to distinguish the PA signal of PACDx from background due to blood. We observed no change in the PA intensity for the control group (PA_Final/PA_Initial = 0.99 ± 0.16) (Fig. 6g,h). However, the ratio in the tumour region was 3.17 ± 0.91, indicating that PACDx had successfully detected elevated GSH in an orthotopic lung cancer model.

Beyond the primary tumour, we also wanted to evaluate PACDx in a liver metastasis model because ~40% of lung tumours have metastasized by the time of diagnosis, with liver being the most common metastatic site⁴⁵. Unfortunately, metastasis was inconsistent using the orthotopic lung cancer model described above. As such, we generated a heterotopic liver cancer model using A549 luciferase cells to test PACDx. When the tumours were imaged via cross-sectional PA imaging, we did not observe any notable changes in the PA intensity for the control group (PA_Final/PA_Initial = 0.94 ± 0.19), but noted a significant turn-on response in the tumour region (PA_Final/PA_Initial = 3.03 ± 0.97) (Supplementary Fig. 32a–c). This key experiment further demonstrates the broad utility of PACDx.

Discussion

The identification of disease biomarkers and the subsequent use of this information to guide therapeutic decision making is highly desirable. In this study, we have utilized a physical organic approach to guide the development of a PA imaging-based CDx for the detection of GSH in murine models of lung cancer. In contrast to commercial probes (for example, GSH Red, λ_max = 580 nm)⁴⁶, PACDx absorbs maximally in the NIR window, and this property facilitates access to deeper regions of the body. Moreover, it is noteworthy that GSH is typically not considered to be an ideal cancer biomarker because current sensing strategies are too insensitive to distinguish between normal and pathological levels, especially in vivo. For example, 5 μM GSH is sufficient to fully activate 2.5 μM GSH Red within 1 h. To overcome this challenge, we performed precision tuning to optimize the S_NAr reactivity, which enabled us to reliably differentiate GSH in the 0.1–10 mM range within the same timeframe. The potential advantage of employing an imaging-based approach rather than a conventional in vitro testing strategy is the potential to visualize changes that occur during disease progression in real time.

We also developed a highly effective prodrug that utilizes the same GSH-mediated activation mechanism to selectively release gemcitabine from a PA imaging dye. Although we could have directly installed the new trigger onto gemcitabine, the ability to perform PA imaging provides us with a powerful handle to monitor drug delivery. Moreover, we did not want to drastically alter the cellular uptake or biodistribution, because the HD dye scaffold has been shown to target tumours, presumably due to the intrinsic positive charge. We envision further exploiting this robust design to selectively deliver other cancer drugs to tumours. With regards to gemcitabine, it is an FDA-approved drug and one of the first-line treatment options for lung cancer. Unfortunately, it is characterized by rapid metabolism, poor bioavailability and low tumour uptake. As a result, frequent doses must be administered to ensure a therapeutic response, and this can lead to severe liver damage. Because PARx is only activated in lung cancer cells, we were able to demonstrate through different dosing regimens (every 7 days versus every 3 days for 21 days) that PARx did not display any off-target toxicity.

At the onset of this work, our primary objective was to develop a PA imaging-based CDx that could identify subjects with lung cancer based on elevated GSH levels. An additional goal was to employ the CDx to identify patients with cancer who would respond to treatment versus those that would not. We were able to design an unprecedented blind study to evaluate the performance of PACDx in the context of these two aims. Through the application of PACDx, we were able to stratify seven mice bearing different tumour types into two groups. Subsequent treatment with the prodrug demonstrated that the group suspected of having lung cancer did indeed respond. When the identity of each animal was revealed, we were pleased to find that we could accurately distinguish between different cancer types based on PA imaging. This study showcases the potential use of PA imaging to detect critical biomarkers and has allowed us to predict—before treatment—which group of animals will have the greatest therapeutic efficacy when treated with PARx.

Owing to the versatile nature of PACDx, we envision that it will find utility in a broad range of important biomedical applications. For example, we have already demonstrated that PACDx can be employed as a fluorogenic probe to detect elevated GSH in live cells using confocal microscopy. Although many fluorescent-based GSH probes exist^47–50, we envision that the optimized chemical reactivity and NIR capabilities of PACDx will make it ideal for the detection and tracking of GSH in patient-derived tumour organoids, as well as patient-derived xenografts via optical imaging. The ability to choose between imaging modes (optical versus PA), will present a user with maximal flexibility when designing an experiment. Because PACDx was optimized to discriminate high GSH levels from physiological concentrations, we were able to use it in our study to detect lung cancer. This was demonstrated using PA imaging in multiple animal models of lung cancer, including a subcutaneous xenograft model, an intrahepatic model and an orthotopic lung model. Moreover, beyond targeting lung tumours, we anticipate PACDx and PARx can be employed to detect other conditions and tumour types featuring elevated GSH levels. Finally, we envision that PACDx can be used to aid drug development efforts, because PA imaging can help to confirm and quantify delivery in real time. Beyond preclinical applications, the potential of our new approach in the clinic will be contingent on the availability of instrumentation that facilitates imaging beyond several centimetres. Because PA imaging is a ‘light-in, sound-out’ technique, sufficient light must reach an imaging probe before a signal can be generated. This is one of the major obstacles of translating this technology. However, the development of endoscopic PA set-ups can overcome this challenge, because the region of interest can be illuminated from within the body, bypassing tissue that can attenuate incident light.

Data availability

All data are available within the Article and its Supplementary Information. Alternatively, data are available upon request from the corresponding author.