Correlation of in vivo tumor response and singlet oxygen luminescence detection in mTHPC-mediated photodynamic therapy

Brian C. Wilson*, Michael S. Patterson, Buhong Li and Mark T. Jarvi* *Department of Medical Biophysics University of Toronto/University Health Network Toronto, Ontario M5G 1L7, Canada Juravinski Cancer Centre and McMaster University Hamilton, Ontario L8V 5C2, Canada MOE Key Laboratory of OptoElectronic Science and Technology for Medicine Fujian Provincial Key Laboratory for Photonics Technology Fujian Normal University, Fuzhou, Fujian 350007, P. R. China wilson@uhnres.utoronto.ca


Introduction
The excited singlet state of oxygen ( 1 O 2 ), is believed to be the main cytotoxic reactive species generated during photodynamic therapy (PDT) for a number of photosensitizers used clinically (porphyrin-and chlorin-based) and for some investigational new agents. 1The focus of the present work is on singlet oxygen dosimetry, so that it is relevant only to this class of photosensitizers and not, for example, to other compounds like palladium bacteriopheophorbides 2 (TOOKAD, WST11) that are photodynamically potent but operate through photo-induced electron transfer leading to generation of reactive oxygen species such as hydroxyl radicals. 2,3 1O 2 is produced in a Type II reaction, in which the excited singlet state of the photosensitizer generated upon photon absorption by the ground-state photosensitizer molecule undergoes intersystem crossing to a long-lived triplet state.This state can then exchange energy with the triplet ground state of molecular oxygen ( 3 O 2 ). 1 O 2 can decay radiatively, emitting near infrared (NIR) luminescence at around 1270 nm.However, 1 O 2 is generally believed to have a very short lifetime (likely ( 1 s) in cells and tissues, 4,5 due to its high reactivity with biomolecules, so that the luminescence signal is very weak ($ 1 : 10 8 probability).In addition, 1270 nm is not in a favorable range for e±cient photodetection using standard devices.Nevertheless, it is technically feasible using nanosecond laser pulses and an extended-wavelength photomultiplier tube (PMT) operating in timecorrelated, single-photon-counting (TCSPC) mode.As reviewed by Jarvi et al.,4 this so-called 1 O 2 luminescence dosimetry (SOLD) technique is now generally accepted as the \gold standard" for PDT dosimetry and has been used as such in a number of studies, for example in characterization of novel photosensitizing agents, including activatable molecular beacons 6 and nanoparticle-based sensitizers. 7ts validity has been demonstrated clearly in cells in vitro, where it has been shown to generate a \universal response curve" of cell killing vs cumulative 1 O 2 counts generated during treatment. 8Importantly, this response curve is independent of the individual treatment parameters such as photosensitizer concentration, light dose and oxygenation.Its validity has also been demonstrated in normal skin models 9,10 and in normal human skin using a timeintegrated technique, 11,12 although the latter may have limitations due to interference from background tissue and photosensitizer °uorescence and/or phosphorescence.SOLD has been demonstrated in a brain tumor model in vivo by Yamamoto et al., 13 but without applying spectral scanning to remove background signals.Recently, Schlothauer et al. 14 investigated the 1 O 2 luminescence generated from topically applied photosensitizer in pig ear skin in vitro, and the data showed that the 1 O 2 kinetics coincides with the microarchitecture of epidermis, such as in ¯ssures and hair follicles.In order to predict the clinical phototoxic response (erythema) resulting from PDT with aminolevulinic acid-induced protoporphyrin IX, Mallidi et al. 12 compared discrete photosensitizer °uorescence-based metrics with corresponding 1 O 2 luminescence-based metrics.They suggested that, at least for this photosensitizer, a dose metric based on its °uorescence photobleaching may be adequate to predict the PDT outcome.While this technique is more clinically applicable at present than SOLD, Jarvi et al. 15 have shown that it may not work for all photosensitizers, depending on parameters such as the level of oxygenation.
Given that there are a number of both technical and conceptual reasons why measuring the volumeaveraged 1 O 2 concentration may fail in the complex and heterogeneous milieu of solid tumors, 16 the objective of this study was to test, in a well-controlled tumor model in vivo and using the most robust spectrally resolved TCSPC technique, whether SOLD does correlate with tumor response.Importantly, to our knowledge this is the ¯rst study showing correlation of in vivo tumor response with the 1 O 2 generated during treatment using a SOLD technique that incorporates both spectral discrimination and time-resolved single photon counting to ensure that the NIR photons detected are from 1 O 2 decay only and are not potentially confounded by unknown background contributions.

Animal model
The well-established dorsal skin-fold window chamber tumor model was used, as shown in Fig. 1(a).Brie°y, window chambers were implanted on female NCRNu mice (22-28 g) anesthetized by intraperitoneal (i.p.) injection of Xylazine (10 mg/kg) and Ketamine (80 mg/kg).For this a 10 mm diameter circular incision was made and the skin from one side of the dorsal skin fold was removed.A titanium chamber was fastened with three screws at the site of incision.A circular glass coverslip was positioned over the opening and clamped to the chamber with a retaining ring.These chambers typically remain viable for several weeks, with the animal housed under normal conditions.9L gliosarcoma tumor cells were transduced via lentivirus vectors to express both green °uorescent protein (GFP) and luciferase (luc) and a highly expressing clone was selected based on the GFP °uorescence signal and the luciferase-mediated bioluminescence signal.These transduced cells, 9L lucÀGFP , were passaged in Dulbecco's modi¯ed medium supplemented with 10% fetal bovine serum and grown to con°uence at 37 C under 5% CO 2 .Cells were removed with trypsin, centrifuged and re-suspended in fresh media at a density of 1:5 Â 10 7 cells/mL for implantation on the same day the chambers were installed.The coverslip was temporarily removed and 8 L of tumor cell suspension was injected into the fascia of the subcutaneous skin layer opposite the window.The injection site was standardized to the upper center of the chamber near to a large vessel.

PDT treatment
The mice were treated 7-8 days after tumor implantation.About 4 h prior to light delivery, meso-tetrahydroxyphenylchlorin (mTHPC: Biolitec, Vienna, Austria) in clinical formulation and dilution 1:20 in 40% ethanol and 60% polypropylene glycol was injected i.p.Although it has a number of limitations, mTHPC was selected as a model photosensitizer here, since it is clinically approved (in Europe for treatment of head and neck cancers) and has been used in previous 1 O 2 dosimetry studies. 15bout 4 h mice were re-anesthetized and restrained on a heated stage.Each mouse received a ¯xed radiant exposure of 12 Jcm À2 at an irradiance of either 35 or 100 mW cm À2 from a 523 nm diode laser over a 5 mm diameter spot centered on the tumor, as visualized by the GFP °uorescence.The treatment and control cohorts were: 35 mW cm À2 and 0.6 mg/kg mTHPC (n ¼ 12); 100 mW cm À2 and 0.6 mg/kg mTHPC (n ¼ 9); 35 mW cm À2 lightonly controls (n ¼ 3) and 100 mW cm À2 light-only controls (n ¼ 3).After treatment, the mice were recovered and returned to normal housing under subdued lighting.

SOLD and fluorescence imaging/spectroscopy
NIR luminescence was detected using the system shown in Fig. 1(b), the basis of which has been described in detail previously. 4,8Brie°y, a frequencydoubled Nd:YLF laser (QG-523-500; CrystaLaser In vivo tumor response to singlet oxygen 1540006-3 Inc., Reno, NV, USA) generated $ 10 ns pulses of 523 nm light at a pulse repetition rate of $ 10 kHz and average power of < 200 mW.This was expanded by a lens to a uniform (top hat) 5 mm diameter spot at the window chamber.The light collected from the tumor passed through a 5position ¯lter wheel that sampled the NIR luminescence spectrum at 1212, 1240, 1272, 1304 and 1332 nm (20 nm FWHM) before being passed to a PMT with extended NIR sensitivity (R5509-14: Hamamatsu Corp., Bridgewater, NJ, USA).The PMT was connected to a TCSPC system triggered by the laser pulse.A small silver-coated prism on the detection optical axis re°ected part of light through 600 nm long-pass and 650 AE 50 nm bandpass ¯lters to an intensi¯ed CCD camera for simultaneous imaging of the white light and GFP °uorescence.

Bioluminescence imaging
Bioluminescence images (BLI) and tumor-cell GFP °uorescence images were collected under general anesthesia to evaluate the tumor response to treatment.In the present analysis, the GFP imaging was used simply as a cross check that the BLI properly reported the viable tumor.At 1 week following tumor implantation the mice were injected with 100 L (150 mg/kg) of D-luciferin dissolved in freshly prepared phosphate bu®ered saline.The mice (typically 5 at a time) were then placed in a whole-body imaging system (IVIS: Perkin Elmer, Waltham, MA, USA).Starting at 2 min post injection, BLI were acquired every 1 min up to 20 min and then every 2 min up to 34 min to track the kinetics of the bioluminescence signal.In addition, a GFP image ( ex ¼ 445-490 nm, em ¼ 515-575 nm) was acquired at the 2.5, 7.5, 20 and 33 min time points.This imaging procedure was performed 5 times for each cohort of mice: 48 h pre-treatment; 4 h pre-treatment, immediately before photosensitizer injection; 2, 5 and 9 d post treatment.Additionally, one cohort (n ¼ 5) was imaged daily for the ¯rst 8 d post tumor implantation to evaluate the tumor growth kinetics.In each animal, the ratio, R ¼ C post =C pre , was calculated, where C pre represents the total BLI counts over the tumor immediately before PDT and C post represents the total counts at a speci¯c time point following treatment.At each post-treatment time point, this ratio was then averaged over all animals.

Results
Response to PDT treatment was based on the changes in the BLI counts integrated over the whole tumor.The average ratios for each post-treatment time point were then classi¯ed into \no response" (R > 1: continued tumor growth), \moderate response" (R ¼ 0:5-1: tumor stasis or slight reduction) or \strong response" (R < 0:5: at least 2-fold reduction in viable tumor cells).Such nonparametric classi¯cation of PDT response has been reported by Ascencio et al. 17 and was applied here in order to address the high tumor-to-tumor variability seen both in the BLI images and SOLD counts, and Fig. 2 presents examples of each response, at a particular time (5 d) post-treatment.
For the \strong response" case, there is little remaining bioluminescence signal after treatment.Conversely, in the \no response" tumors the BLI images before and after treatment showed little change, similar to the light-only controls.We note that the fact that the BLI showed little change in the control (light-only) animals at 5 d post treatment suggests that the tumor may have reached its growth limit by the time of treatment, i.e., at 7-8 days following implantation: this is not unexpected in the chamber model where the tumor is physically con¯ned on one side by the window and on the opposite side by the overlying skin.
Quantitative bioluminescence imaging has also been used in previous studies of PDT treatment response 18 and was further validated here [Fig.3(a)] by comparison with the observed changes in the GFP signal that also reports functionally active tumor cells.Figures 3(b) and 3(c) then summarize the treatment response results, plotted in two di®erent ways.Figure 3(b) shows how the cumulative 1 O 2 luminescence counts in individual animals, as well as the averages over all animals, di®ers between the response groups.Despite the high level of scatter in the data, which is presumably related to the variability in mTHPC uptake and tissue oxygenation in the heterogeneous tumors, there is a clear trend to higher 1 O 2 levels in those tumors showing the greatest level of cell kill: this was statistically signi¯cant (Student's t-test) between the \strong-response" group and both the \control" and \no response" groups.Plotting the data as a more conventional response curve in Fig. 3(c) shows again the high scatter between tumors, but there is a clear response-dose relationship consistent with a sigmoidal response, as might be expected.

Discussion and Conclusions
This study has demonstrated the validity of 1 O 2 luminescence as a valid PDT dose metric in an in vivo tumor model.This extends work in photosensitizer solution, tumor cells in vitro and normal tissue (skin) in vivo, reported previously by ourselves 8,9 and by other groups [10][11][12]14 using variations on the SOLD instrumentation, measurement technique and biological models. Th most direct comparison can be made with the in vivo study by Yamamoto et al., 13 in which 1 O 2 luminescence dosimetry was used in a gliosarcoma tumor xenograft model, assessing the PDT response by quantitative histopathology.They showed that, for a given total light °uence, both the 1 O 2 counts and the tumor response decreased as the °uence rate increased (from 30 to 120 mWcm À1 using an interstitial diffusing ¯ber to deliver the light).This reduction in PDT e±cacy at high light °uence rates has been reported also in vitro 19 and in vivo 20 and is well understood in terms of photochemical depletion of ground-state molecular oxygen.21 A particular advantage of the luminescence instrument used by Yamamoto et al. 13 was the ability to collect the luminescence emitted from the tumor by a ¯ber optic bundle.However, the signal was neither spectrally resolved nor time-gated to remove the prompt NIR signal (<$ 1 s after the excitation pulse) so that, as discussed by Jarvi et al. 15 and Lin et al., 22 there may be signi¯cant contributions from background °uorescence and/or phosphorescence.Hence, we believe that the present work represents the ¯rst demonstration of the correlation between in vivo tumor response and 1 O 2 using a rigorously validated SOLD system.
We note that, as discussed elsewhere, 23 measurements done at the same time in these experiments using speckle-variance optical coherence tomography to image changes in the tumor microvasculature also demonstrated good correlation with 1 O 2 generation.Hence, there is reason to believe that volume-averaged 1 O 2 dosimetry is a valid metric for vascular as well as cellular responses, at least using this particular photosensitizer.
These results also con¯rm the value of SOLD as a gold standard comparator for indirect methods, such as the use of secondary °uorescent reporter molecules 24 whose spectral characteristics change upon chemical interaction with 1 O 2 .They also motivate the development of new technologies for this purpose, including NIR-sensitive imaging arrays 25 and superconducting nanowire photodetectors that have much higher 1270 nm quantum e±ciency than the best PMTs and that have recently been demonstrated for 1 O 2 luminescence detection through single optical ¯bers. 26Both advances should facilitate the translation of SOLD into clinical applications, where a robust method for monitoring the true administered PDT \dose" in patients currently impedes the adoption of this modality.This is particularly true in oncology, where the tumor responses can be highly variable even for the same applied photosensitizer and light doses. 1,27ne of the limitations of the present study was that the use of the window chamber, which has the advantage of providing direct and quantitative assessment of the tumor response to PDT, meant that it was not possible to determine if the long-term treatment response also correlates well with the SOLD measurement, since the chamber preparation only remains viable for a few weeks.Clearly, this is of the greatest interest for potential clinical applications.Hence, in the next phase of this work one promising option is to use the aforementioned ¯bercoupled nanowire detectors 25 for interstitial SOLD measurements so that a subcutaneous or even an orthotopic tumor model can be used in order to follow the responses over a much longer period.

Fig. 1 .
Fig. 1.Experimental setup.(a) Window chamber model shown in place on the imaging stage.(b) SOLD system: PMT, photomultiplier tube; CCD, intensi¯ed charge coupled device camera for °uorescence imaging; PC, personal computer.Note that the data from the °uorescence spectrometer are not included in the present analysis.

Fig. 2 .
Fig. 2. Representative BLI for the di®erent levels of treatment response at 5 d post treatment.The green signal is the falsecolor intensity of the BLI signal.The ring in the gray-scale photograph in each case delineates the transparent window within which the bioluminescence (and °uorescence) images are collected.

Fig. 3 .
Fig. 3. (a) Correlation between GFP and bioluminescence imaging, plotting in each case the ratio of the integrated image signal before and at 5 d after PDT treatment: each point corresponds to an individual tumor.(b) Cumulative luminescence counts in individual tumors (open symbols) and averaged over all tumors (solid symbols) for each of the BLI treatment response groups.Note that the negative data points are the results of noise in the background subtraction at low 1 O 2 signal levels.(c) Sigmoidal BLI response plot, using the initial 1 O 2 luminescence counts at the start of PDT treatment as the dose metric, which is an alternative to the cumulative SOLD counts used in (a).