Detection of singlet oxygen luminescence for experimental corneal rose bengal photodynamic antimicrobial therapy

: Rose bengal photodynamic antimicrobial therapy (RB-PDAT) treats corneal infection by activating rose bengal (RB) with green light to produce singlet oxygen ( 1 O 2 ). Singlet oxygen dosimetry can help optimize treatment parameters. We present a 1 O 2 dosimeter for detection of 1 O 2 generated during experimental RB-PDAT. The system uses a 520 nm laser and an InGaAs photoreceiver with bandpass filters to detect 1 O 2 luminescence during irradiation. The system was validated in RB solutions and ex vivo in human donor eyes. The results demonstrate the feasibility of 1 O 2 dosimetry in an experimental model of RB-PDAT in the cornea.


Introduction
Corneal infection (keratitis) is a common ocular emergency which can lead to permanent damage to the cornea and possible blindness and complete loss of the eye due to endophthalmitis [1]. There are various antimicrobial drugs to fight these infections, but they often have limited success treating resistant or atypical organisms [2,3].
Preliminary studies demonstrate the potential of Rose Bengal Photodynamic Antimicrobial Therapy (RB-PDAT) as a therapy for treating microbial keratitis [4][5][6][7][8]. The infected portion of the cornea is treated with the photosensitive dye Rose Bengal (RB), then exposed to 500-560 nm light. Reactive singlet oxygen ( 1 O 2 ) generated by this process destroys the infecting microorganisms [9,10]. While preliminary results have been promising (>80% of patients treated with experimental RB-PDAT avoid emergency cornea transplant), a better understanding of the fundamental photochemical processes for generating 1 O 2 is required to maximize treatment efficacy [8]. The goals of the optimization include maximizing 1 O 2 production, maximizing treatment depth, and minimizing treatment time [11,12]. To make this possible, an accurate measure of 1 O 2 dose must be established. 1 O 2 dose can be quantified by measuring 1270 nm luminescence [13,14]. However, this method is technically challenging within biological environments due to extremely weak emission [15][16][17][18][19][20][21]. As a result, highly sensitive detection systems, sometimes combined with time-gating, are required [13,[22][23][24][25][26][27][28][29][30]. Due to their complexity it would be challenging to integrate these detectors into a clinical system adapted toward 1 O 2 measurements in the human cornea.
Recently, dosimetry systems have been developed using InGaAs cameras and photoreceivers, which are able to measure real-time continuous wave emission at sensitivity levels approaching that of photomultiplier tubes [13,[22][23][24][31][32][33][34][35]. While InGaAs avalanche diodes might still use time-gating, most of these systems perform continuous wave measurements, relying purely on spectral filtering to select for 1 O 2 luminescence at approximately 1270 nm. Several of these devices have successfully measured 1 O 2 production in the skin in vivo [22,36]. To date however, no system has been specifically built for the purposes of 1 O 2 measurement in the corneal surface during experimental RB-PDAT.
In this work we present a proof-of-concept system for corneal 1 O 2 measurement. This study relies on methods demonstrated in previous work for animal models, which use continuouswave photosensitizer excitation and 1 O 2 measurement using spectral filtering with an InGaAs photodetector [36]. While these works focus primarily on measurements of skin or superficial tumors, our 1 O 2 dosimeter is focused specifically on 1 O 2 luminescence measurement during experimental corneal RB-PDAT. The system generates 1 O 2 using a continuous-wave 520 nm laser excitation source to stimulate RB solutions and human donor eyes treated with RB photosensitizer. Optical filters provide spectral separation between collected 1 O 2 luminescence (approximately 1270 nm) and background fluorescence from RB and/or tissue autofluorescence. The goal of this study was to demonstrate feasibility of this approach for corneal surface 1 O 2 luminescence dosimetry.

Description of a singlet oxygen 1 O 2 dosimeter
A 520 nm, 4.5 mW diode laser emitting a 4.6 mm x 1.7 mm elliptical beam (CPS520, Thorlabs) was used. Laser emission is controlled manually using a shutter (Fig. 1). Preliminary testing of the laser showed some residual IR emission. Therefore, the laser light is filtered to remove residual IR emission using a glass filter (FGS900-A, Thorlabs). The laser light passes through a dichroic shortpass filter (#86-697, Edmund Optics). This dichroic filter transmits the incident green laser beam and reflects light with wavelengths above 1155 nm toward the detector, which includes 1 O 2 luminescence (approximately 1270 nm). The laser is focused with a 20x objective lens (W.D. = 3.2 mm, N.A. = 0.4, f = 9.0 mm), onto the sample (Fig. 1, Lens 1). The calculated spot size of the beam is 3.5 µm × 1.3 µm, assuming that the focused beam is Gaussian and diffraction limited. The corresponding peak irradiance, not accounting for transmission losses, is 2.5 × 10 5 W/cm 2 . The near infrared radiation emitted by the sample is collected by the same lens and reflected by the dichroic filter through a sliding filter holder. In order to spectrally separate 1 O 2 luminescence (approximately 1270 nm) from RB fluorescence and/or tissue autofluorescence and observe a 1 O 2 luminescence peak, four bandpass filters were accommodated in the sliding filter holder (1200, 1250, 1277, and 1300 nm, with FWHM of 10, 10, 20, and 12 nm, respectively. All filters were purchased from Edmund Optics, except the 1277 nm filter, which was purchased from Omega Optical). The 1277 nm bandpass filter selects for 1 O 2 luminescence, while the other bandpass filters (1200, 1250, and 1300 nm) provide a measurement for RB fluorescence and/or tissue autofluorescence within this range. The 1277 nm filter was chosen as the closest off-the-shelf available filter for 1 O 2 luminescence measurement. Since the 1 O 2 luminescence is a weak signal that may be masked by these intrinsic fluorescence sources, two 1150 nm longpass filters are stacked in the detection path. The light transmitted by the filters is then focused with a plano-convex lens (Lens 2, f = 8 mm, ACL12708U-C-SP, Thorlabs) onto an InGaAs photoreceiver with femotwatt sensitivity (PDF10C, Thorlabs). To ensure proper alignment of the optics and maximize signal detection, the dichroic filter is placed on a stage capable of fine rotation up to ± 7°, and both the collecting lens and focusing lens have been mounted such that their axial positions are adjustable by ± 1.5 mm.
The InGaAs photoreceiver was connected through a 75 Ω BNC cable to a resistance terminator (10 kΩ) to prevent RF signal reflection and reduce noise, and then to an oscilloscope (PicoScope 2204A, Pico Technology). Finally, the oscilloscope signal is recorded with a computer and processed in MATLAB 2019a (MathWorks, Natick, MA, USA).

Data collection and processing
In all experiments, the 1 O 2 luminescence is measured using the same procedure. At least 15 minutes before any measurement, with the shutter closed, the 520 nm laser and the InGaAs photoreceiver are turned on to allow time for warm-up as recommended by the manufacturers. The 1277 nm bandpass filter is placed in the detection arm. The sample is mounted onto the optical bench in front of Lens 1. The room lights are then turned off and the laser shutter is opened. The dosimeter is precisely positioned in front of the sample using the XYZ linear translation stage until the signal detected is maximized. The background signal recorded for two seconds. The shutter is then opened, and the recording continues for approximately 15 seconds (Fig. 2). The 1277 nm bandpass filter is then removed and replaced with one of the other filters. This procedure is repeated three times for each bandpass filter. The oscilloscope records data at 750 Hz.
The recorded signal is filtered using a 50 Hz low-pass filter to remove 60 Hz AC noise. Immediately after the laser shutter is opened, some of the measured signals include an initial decay phase (Fig. 2). This initial decay phase is most obvious in measurements at 1277 nm and is thought to be due to 1 O 2 luminescence decrease due to decreased oxygen supply within the tissue [37]. To have a more reliable comparison between measurements, only the data after the decay phase, from 3.5 to 16.5 s, were used in the analysis. The averaged background signal from each recording is subtracted from the average signal measurement. These data are then corrected for the transmission provided by the manufacturer for each bandpass filter.

Measurements of 1 O 2 luminescence in solution
In order to verify if the dosimeter can measure 1 O 2 luminescence, we first tested the device on solutions of Rose Bengal contained in a 1 cm optical cuvette mounted in a custom designed holder. The cuvette is held approximately at a 25°angle to prevent back reflection of the laser generated by the window surface into the objective lens (see Fig. 3(a)). XYZ position of the dosimeter in front of the cuvette was set by finding where the signal was maximum using the 1277 nm BP filter with a solution of 0.1% RB in distilled water (18 MΩ-cm resistivity, NERL High Purity Water, Thermo Scientific). We performed five experiments measuring 1 O 2 luminescence using triplicate measurements. 4. 99% D 2 O: to provide a measurement of how much of the signal was due to back reflection of the 520 nm laser and confirm no 1277 nm peak is observed when RB is not present in solution.
5. Distilled water: to provide a measurement of back reflection of the 520 nm laser and confirm no 1277 nm peak is observed when RB is not present in solution.
In Fig. 3(b), the peak at 1277 nm closely matches the reference 1 O 2 spectrum for the 0.1% RB in D 2 O solution [40]. Combined with the increased peak using D 2 O and the decreased peak from NaN 3 , these results show that the dosimeter is measuring 1 O 2 . The signal that remains for the 0.1% RB in 200 mM NaN 3 solution after quenching is assumed to be background signal due to RB fluorescence [36]. Subtracting this RB fluorescence background signal from the 1277 nm signal for the 0.1% RB in distilled water and 0.1% RB in D 2 O solutions, we estimate that 79% and 92% of the 1277 nm signal can be attributed to 1 O 2 luminescence, respectively. Note that 0.1% RB corresponds to 1.03 mM RB. Three measurements were taken using bandpass filters at 1200, 1250, 1277, and 1300 nm. Normalized average dosimeter voltage signal is displayed, where 100% corresponds to 293.4 mV or 3.99 × 10 12 1 O 2 molecules per second (calculated, see appendix). In addition, a sample 1 O 2 spectrum is displayed for comparison [40]. Note that all data points have error bars, however almost all are too small to be visible beyond the data point marker.

Effect of RB concentration on 1 O 2 signal
We sought to determine if the system could successfully record 1 O 2 luminescence signals for a range of RB concentrations that are relevant to clinical use. In order to test this, we performed an in vitro experiment comparing dosimeter signals with varying RB concentrations in distilled water in a glass cuvette as described in the previous section: 0%, 0.05%, 0.10%, 0.15%, and 0.20%. These concentrations were chosen because this is the RB concentration range that has been shown to have clinically relevant antimicrobial effect [4,5,8,41,42]. Prior studies have demonstrated that lower concentrations do not produce an efficient antimicrobial effect. The setup and measurement were the same as shown in Fig. 3(a).
Normalized 1 O 2 dosimeter signal is shown in Fig. 4. Signal at 1277 nm decreased with increasing RB concentration. All RB solutions had highest signal at 1277 nm compared to other measured wavelengths. 1 O 2 dosimeter signal likely decreased with increasing RB concentration, which is explained by the formation of aggregates at high concentration of RB inducing selfquenching [43][44][45][46]. It is likely that the signal at 1277 nm would have increased at concentrations of RB lower than 0.05%, and that 1277 nm signal would show a positive trend at RB concentrations below 0.015% (150 µM), above which self-quenching RB multimers are thought to begin forming in solution [43]. Attenuation of the 1277 nm light directly due to increasing RB concentration is unlikely to be a contributing factor because RB is nearly transparent to 1277 nm light even at concentrations as high as 10 wt% RB [47].

Purpose and methods
To determine if the 1 O 2 dosimeter can detect 1 O 2 luminescence during RB-PDAT in the cornea, we tested the 1 O 2 dosimeter using five human donor eye pairs (ten eyes) treated with solutions of RB. See Table 1 for a summary of different treatment conditions.

Table 1. Treatment Conditions for Whole Globe Experiment
All measurements were performed in triplicate at the same XYZ location on the cornea. The system collects luminescence from the corneal surface through a corneal depth of approximately 150-200 µm. The XYZ location was stimulated for approximately 3 to 4 minutes in total in order minimize the effects of photobleaching. Due to the manual operation of the system, each triplicate measurement required approximately 30 minutes including positioning and alignment of the eye and irradiation. Despite possible non-uniformity of corneal RB distribution, multiple XYZ locations were not used due to concerns that excessive measurement time would become a confounding variable.
1. Group 1 -no treatment (3 eyes): serves as a baseline measurement of how much of the signal corresponds to back reflection of the 520 nm laser and/or tissue autofluorescence.
2. Group 2 -20% dextran treatment (7 eyes): serves as a baseline measurement of how much of the signal corresponds to back reflection of the 520 nm laser and/or tissue autofluorescence, as well as how much dextran affects measured signal.  [15]. Oscilloscope data acquisition was adjusted to 300 Hz for this measurement. To confirm that the decrease in power was not caused by the laser instability, we performed a separate measurement of the laser power stability for 45 minutes and measured 0.27% power variation during that window. This confirms that that laser instability did not play a role in signal decrease during this experiment.
Corneal preparation: Donor eyes not suitable for clinical use were graciously provided by the Florida Lions Eye Bank located at Bascom Palmer Eye Institute. Eyes were used between 4 and 18 days after donor death, with a median eye post-mortem time of 10 days. Eyes were cleaned of extraocular tissue and then placed in a custom 3D printed vacuum holder to allow for easy handling (Fig. 5). The epithelium in all corneas was debrided (de-epithelialized) to a diameter of 8 mm. To restore a normal intraocular pressure (IOP), 0.9% NaCl (1-1.5 mL) was injected into the vitreous humor through the optic nerve using a 27 G needle. To both clarify the cornea and bring it to a physiologic thickness, a small volume (approximately 50 µL, or until cornea was taut) of 20% dextran solution was injected into the anterior chamber using a 25 G needle.
Like the corresponding RB solution, the 20% dextran solution was prepared using approximately 40 kDa dextran (D1662, Sigma-Aldrich) in a solvent of either D 2 O, 200 mM NaN 3 , or distilled water. The eye was then placed face down in a pool of 20% dextran solution for 60 minutes. A custom eye holder with a cylindrical well filled with 20% dextran solution was created to ensure only the cornea was exposed to dextran. To verify that the corneal thickness and pressure reached physiologic values, triplicate pachymetry (Accutome PachPen Pachymeter 24-510) and tonometry (Clement-Clarke Perkins MK2 Handheld Tonometer) measurements of the donor eyes were performed, respectively, after removal of epithelium and after 60 minutes of 20% dextran treatment. After all preparation steps, corneal thickness was measured as less than 599 µm for all but 3 eyes (one eye each in groups 3-5), and IOP ranged from 0-74 mmHg for all eyes. Low IOP in two eyes (one each in groups 3 and 5) resulted from leakage from the tissue surface after the initial saline injection. However, IOP should not have significant effect on luminescence measurements because it is the 1 O 2 generated by RB, not the corneal tissue, which produces the luminescence. To apply RB solution, a 6 mm corneal sponge was presoaked with RB, then placed on the corneal surface. RB was applied for 30 minutes, with three drops added every 3 minutes using a 10 G pipette. After RB treatment, the corneas were washed with 0.9% NaCl.

O 2 measurement:
Similar to the in vitro experiments, whole globes were first placed in a custom holder and mounted in the system. Alignment was accomplished by adjusting XYZ position to the point of maximum signal in the RB stained region of the eye using the 1277 nm bandpass filter. All eyes were measured three times with each bandpass filter (1200, 1250, 1277, and 1300 nm). All eyes from group 1 were measured before any treatment. In group 2, all eyes were measured after 20% dextran treatment. Groups 3, 4, and 5 were measured after RB treatment.
Photobleaching measurement: After 1 O 2 measurements using all four bandpass filters, 1 O 2 dosimeter signal was measured at a new RB stained section of the cornea. The signal was measured at 1277 nm over a 16-minute period for group 3 (treated with 0.1% RB in D 2 O) sampled at 300 Hz. Hydration of the cornea was not maintained during this period, which may have resulted in decreased signal due to changes in tissue properties and corneal thickness [48,49]. Because there is an initial spike in the measured signal (see Fig. 2), the first recorded 10 seconds of data was removed to more clearly observe photobleaching.

Results and discussion
1 O 2 dosimeter results are shown in Fig. 6. All RB treated groups had an increased 1200 nm signal compared to signal at 1250 nm. Measuring group 3 over a 16-minute period showed a linear decrease in 1277 nm signal for all eyes (Fig. 7).
The spectral measurements from the whole donor eyes showed significantly increased signal at 1277 nm for the 0.1% RB treated groups 3, 4, and 5 compared to groups 1 and 2. Between the RB treated groups, 0.1% RB in D 2 O had the highest mean signal at 1277 nm, followed by 0.1% RB in H 2 O and finally 0.1% RB in NaN 3 . However, these differences were not statistically significant. Quenching by NaN 3 did not decrease luminescence and use of D 2 O did not increase luminescence, possibly due to higher RB concentration in the cornea or rapid quenching of 1 O 2 by corneal collagen (see next section) [39]. RB is strongly associated with collagen in cornea, thus the 1 O 2 is generated very close to the protein and may react with nearby amino acids of the collagen chains. This would shorten the 1 O 2 lifetime, making it less likely to encounter, and be quenched by, azide. Consequently, in cornea, sodium azide may not be an efficient quencher of 1   appears to closely match the 1 O 2 spectrum from Butorina et al. with the exception of the signal at 1200 nm [40]. Despite the failure to significantly quench or boost 1 O 2 signal in corneas, the clear peak at 1277 nm demonstrates that we are successfully measuring 1 O 2 [26].
Although repeat studies and RB fluorescence measurements to determine local RB concentration are necessary, our preliminary results suggest photobleaching appears to progressively reduce the luminescence intensity over the course of 16 minutes. As RB is photobleached, it can no longer generate 1 O 2 and no RB can diffuse to replace the photobleached molecule because the majority of the RB is fixed in place bound to stromal collagen [53]. As a result, the 1 O 2 luminescence signal depletes over time as more RB molecules are continually deactivated. Another possible explanation for these results is that the cornea continued to thin during measurement, causing the corneal surface to gradually move away from the dosimeter focal plane. Extrapolating these data, we can infer that the measurements shown in Fig. 6 were subject to less than 15% drop in signal strength over the 3 to 4 minute measurement window, which is not significant enough to alter the conclusions of this study.

Discussion
To the best of our knowledge, this is the first direct measurement of singlet oxygen ( 1 O 2 ) luminescence in corneal tissues. While this work uses a detection scheme presented in other works, this is the first application to corneal 1 O 2 measurement using Rose Bengal photosensitizer [28][29][30]36]. We present a 1 O 2 dosimeter system that uses an in-line excitation laser and a fixed-gain, room-temperature InGaAs photoreceiver to measure 1 O 2 generated from RB in solution and during experimental RB-PDAT in human donor eyes. Our results indicated a clear luminescence peak near 1277 nm that increases in the presence of a 1 O 2 booster (D 2 O) and decreases in the presence of 1 O 2 quencher (NaN 3 ) in solution. In addition, we were able to characterize 1 O 2 signal from a range of clinically relevant concentrations of Rose Bengal.
In order to use this dosimetry system to measure the dose of 1 O 2 , we must convert the measured luminescence to a quantity of 1 O 2 . As discussed in Appendix 1, for this setup it is calculated that 1 V ≈ 1.36 × 10 10 1 O 2 molecules generated in the measured sample volume. This calculation is useful because, in addition to providing us with a measurement of 1 O 2 content, it also allows us to perform direct comparisons with results from theoretical models for photodynamic therapy, as well as comparison with other dosimeters [25,54,55].
Comparing the ex vivo whole eye measurements to in vitro measurements, we see an increased 1200 nm signal ex vivo. We suspect this is due to the effects of performing these measurements in tissue. It is known that RB binds tightly with collagen, and the resultant local concentration of RB in the corneal stroma is much higher than in solution [39,53]. At the same time, there is potentially shorter 1 O 2 lifetime in collagen matrix compared to in distilled water, so 1 O 2 signal within the cornea is lower for a given RB concentration [15][16][17][18][19][20][21]. Thus, we would expect a large increase in RB fluorescence to be observed even at 1200 nm, with only a moderate change in 1 O 2 signal at 1277 nm. On the other hand, signal increase for all wavelengths for the 20% dextran treated control eye is likely due to decreased scattering, and therefore increased back reflection of 520 nm laser light [48,[56][57][58][59].
However, compared to the in vitro study, there was not a significantly increased signal for the 0.1% RB in D 2 O treated samples compared to 0.1% RB in distilled water. In addition, there was no decrease in 1277 nm signal for the eye treated with NaN 3 . Collagen is known to react with and quench 1 O 2 , so that also is likely contributing to reduced 1 O 2 signal [16]. Because RB is strongly associated with collagen in the cornea, 1 O 2 is produced in proximity to nearby amino acids of the collagen chains. 1 O 2 reacts with these nearby amino acids, making it less likely to interact with NaN 3 or D 2 O. In addition, the local concentration of RB at the corneal surface is much higher than the concentration in solution due to the favorable interaction between RB and collagen [39]. This higher RB concentration would result in higher 1 O 2 production despite the lower measured luminescence signal. This higher 1 O 2 production might be outpacing the rate of quenching by NaN 3 , and thus no significant difference is observed. In addition, a higher RB concentration at the corneal surface may be lowering 1 O 2 luminescence in all studies, both due to attenuation of light, preventing excitation of RB deeper in the cornea, as well as due to self-aggregation behavior of the RB dye [39,43,46].
The human donor eyes in this study were measured between 4-18 days after donor death. This long post-mortem interval likely allowed some collagen breakdown within the cornea while stored at 4°C [60,61]. If so, the collected 1 O 2 luminescence signal might be lowered due to the exposure of more His and Tyr residues which readily react with 1 O 2 , thus reducing the its lifetime [16,[62][63][64][65]. Furthermore, 1 O 2 reaction with ascorbate within the cornea may affect the collected 1 O 2 luminescence [66,67]. The concentration of ascorbate would be expected to decrease due to oxidation during the post-mortem period, thus increasing collected 1 O 2 luminescence. Taken together, we might expect to see small changes in 1 O 2 luminescence from a long post-mortem interval. Finally, higher quality corneas are usually reserved for transplantation. It is possible the quality of the received tissue or the cause of death could have impacted the results. However, none of these effects should change the overall trends observed in the measured results.
The attenuation coefficient of a healthy cornea around 500-550 nm (<10 cm −1 ) is potentially orders of magnitude lower than the attenuation coefficient of 0.1% RB treated corneal tissue (46-2,300 cm −1 , depending on the estimated RB concentration in cornea) [39,53,68,69]. Additionally, in healthy cornea the attenuation coefficient around 1000 nm is <1.4 cm −1 [70]. We therefore do not expect that attenuation of green light or 1 O 2 luminescence by the cornea will have a significant effect on the outcome. Additional attenuation may be present in vivo in a diseased cornea due to increased scattering. Assuming scleral tissue would have similar optical properties to diseased cornea (approximate attenuation at 500 nm is <10 cm −1 and at 1050 nm is <2 cm −1 ), we still expect that the attenuation will be dominated by Rose Bengal given its very high absorption [71,72].
Though the system demonstrated successful 1 O 2 luminescence measurement, a more rigorous study with larger sample size will be required to establish more robust values for 1 O 2 concentrations generated in the cornea. In addition, to reduce differences in measurements due to sample variation, future studies should measure the same eyes before and after treatment in a matched-pair study design. While at least three eyes were measured per treatment group, small sample size was a significant limitation of the study, which is not able to account for variance between donor eyes, in experimental and surgical technique, and heterogeneity of RB distribution on the corneal surface. While some variation has been observed in studies on ex vivo and in vivo rabbit corneas, differences between samples would not be expected to significantly impact outcomes for this proof-of-principle study [39,50,51].
To ensure that there is sufficient concentration of 1 O 2 booster and quencher in the cornea, a higher NaN 3 concentration and a 24 hour corneal soaking period in 20% dextran solution will need to be used in future studies [48,56]. Lastly, it is clear that photobleaching plays a role in these measurements. This aspect of the study needs to be controlled for in future studies by measuring at different locations on the cornea, as well changing around the order of measurements so as not to favor the result from any one bandpass filter.
The current system shows the feasibility for experimental studies on cadaver eyes which will help optimize RB-PDAT treatment parameters. Ideally, this system would be used during clinical RB-PDAT application to monitor 1 O 2 dose. To realize this goal, successful measurement of 1 O 2 luminescence will need to be established using the clinical RB-PDAT light source (525 nm LED source, 6-16 mW/cm 2 ) [8]. This clinical light source has 10 7 times lower irradiance than the focused 520 nm, 4.5 mW laser presented in this study, but this is mitigated to some extent because a much larger volume of the cornea would be irradiated and measured. Experiments will need to be carried out to verify successful 1 O 2 luminescence measurements using the clinical source at these lower irradiance levels. In addition, to bring the system closer to in vivo and clinical use, a longer focal length lens will need to replace the current 20× objective for Lens 1. This will minimize errors due to small movements of the subject. Finally, the system will need to be transferred to a free-standing configuration that can be integrated within the clinical RB-PDAT delivery system so that it can more easily perform measurements on subjects.
A general limitation of the current dosimeter design is that it is only capable of measuring one small spot, failing to capture the entire picture across the cornea. Rose Bengal distribution can vary along the corneal surface, and so a 2D image of corneal 1 O 2 may prove valuable. A near-infrared camera or scanning based system instead of the current fixed-position InGaAs photoreceiver detection might be a better able to obtain a more complete picture of the distribution of 1 O 2 across the corneal surface [22,73].
A final limitation of our dosimeter is that it is unable to provide information on RB-PDAT dose at different corneal depths, which becomes important for deep invading organisms such as Fusarium or Acanthamoeba [11,12].

Conclusion
We present the design of a dosimeter for use in experimental RB-PDAT. Our studies show that the system can be used to measure 1 O 2 luminescence. The device will be used in further studies to optimize treatment parameters. In addition, a next step will be to demonstrate the feasibility of using 1 O 2 luminescence measurements to estimate the efficiency of RB-PDAT.

Appendix 1
The number singlet oxygen molecules produced per second in the measured volume, n1 O 2 (t), is proportional to measured luminescence signal L measured (t) (photons s −1 @ 1277 nm) as described by Eq. (1): where τ L = k L −1 is the luminescence lifetime (τ L = 5.55s for 1 O 2 in water) and f collected is the fraction of generated photons that are collected by the system [55]. While τ L was determined in water, it does not change with differing media [27,55,74].
Calculation of f collected is based on product of the transmission and numerical aperture of the optical system at 1277 nm. Total transmission of the optics at 1277 nm is approximately 69%, and the numerical aperture of Lens 1 = 0.4. The fraction of the total luminescence output collected by Lens 1 is calculated by dividing the solid angle collected by Lens 1 (a function of NA) by the solid angle of the sphere of isotropic 1 O 2 luminescence (4π). For our setup, f collected is approximately 2.9 × 10 −2 . The conversion from signal to 1 O 2 dose assumes that the value of the collection efficiency is constant across experiments. In the present study we used a microscope objective with high numerical aperture (NA = 0.4) to collect the signal. Due to the high numerical aperture, the system will be more sensitive to axial positioning errors, which may produce variations in the collection efficiency. In clinical use, a significantly lower numerical aperture providing a significantly longer depth of focus will be implemented to meet the requirement for a longer working distance. The clinical system will therefore be less sensitive to variations in position and provide a more reliable conversion factor. Finally, L measured (t) can be calculated based on the characteristics of the InGaAs photoreceivers: where L measured (t) is optical power incident on the photoreceiver (photons s −1 ), V OUT is the measured output voltage (V), R(λ) is responsivity of the photoreceiver at a given wavelength (approximately 0.9 A/W @ 1277 nm), G is detector gain (10 11 V/A ± 10%) and hν is the photon energy at 1277 nm (1.56 × 10 −19 J). With these parameters and combining Eq. (1) and (2), we find: n1 O 2 (t) = 1.36 × 10 10 * V OUT Equation (3)