3.1.

Singlet Oxygen Feedback Delayed Fluorescence in Tissue Phantoms

First, the validity of the method was tested on $50\mu \mathrm{M}$ ${\mathrm{AlPcS}}_4$ dissolved in tissue phantoms consisting of 1% intralipid and 1% blood in PBS. Figure 2(a) shows four tubes with the following contents (from right to left): (1) air-saturated sample alone, (2) the sample after removing oxygen by a glucose oxidase/catalase scavenging system,⁴⁶ (3) the sample with addition of 50 mM ${\mathrm{NaN}}_3$, a specific quencher of ${}^1{\mathrm{O}}_2$, and (4) the sample bubbled with oxygen for 5 min. The PF intensity from all the samples has roughly the same intensity and identical lifetime of 6.4 ns. In contrast, DF from individual samples differs strongly. A DF image sequence with $0.5\text{-}\mu \mathrm{s}$ gate width and delays increasing from 0.5 to $8\mu \mathrm{s}$ was captured. The DF kinetics of the air-saturated sample displays a rise–decay shape (with rise time of $0.8\mu \mathrm{s}$ and decay time of $1.7\mu \mathrm{s}$), which is a footprint of the SOFDF mechanism.^20,25 The DF is strongly quenched by ${\mathrm{NaN}}_3$, which proves that SOFDF takes place. The oxygen-saturated sample has a high initial amplitude of SOFDF signal, which however decays quickly ($0.6\mu \mathrm{s}$) due to triplet lifetime shortening by oxygen quenching. The oxygen-depleted sample shows poor intensity, which further supports that DF is oxygen-mediated. The lifetime is prolonged only little, to $2.2\mu \mathrm{s}$, indicating that apart from oxygen quenching there is another significant deactivation pathway for triplets.

Fig. 2

${\mathrm{AlPcS}}_4$ $50\mu \mathrm{M}$ in tissue phantoms in tubes. (a) PF image. From left to right: oxygen-saturated; with addition of ${\mathrm{NaN}}_3$; oxygen-depleted; and air-saturated samples. (b) PF decay kinetics, (c) overall DF emission in gate interval of 0.5 to $8\mu \mathrm{s}$, (d) the first time-gated frame at 0.5 to $1.0\mu \mathrm{s}$, e) the 12th time-gated frame at 6.0 to $6.5\mu \mathrm{s}$, and (f) DF lifetime image calculated from the time-gated sequence of frames using Eq. (3). (g) DF kinetics fitted with two exponentials (black dashed line).

3.2.

Fluorescence of Aluminum(III) Phthalocyanine Tetrasulfonate in a Mouse

Figure 3 displays the PF images of a tumor-bearing mouse that was intravenously injected with ${\mathrm{AlPcS}}_4$. This showed excellent accumulation in the tumor, but it was also present in the abdominal area and to a lesser extent in other tissues including skin, which is common after systemic administration of photosensitizers. The PF lifetime image was calculated from a series of time-gated images with 3 ns gate width and increasing delay. The lifetime of PF is clearly longer in the abdominal area (5.2 ns) than in other parts of the body, including the tumor (4.7 ns), which indicates that the microenvironment and immediate surroundings of ${\mathrm{AlPcS}}_4$ molecules is different, with less quenching in the abdominal area. After delivery of PDT light dose to the tumor, the PF intensity increased and PF lifetime got longer from 4.7 to 5.2 ns, thus reaching the same value as in the abdominal area. This indicates that lysosomal permeabilization in the tumor took place,^37,38,47 releasing ${\mathrm{AlPcS}}_4$ from lysosomes to cytosol and changing the immediate surroundings of ${\mathrm{AlPcS}}_4$ molecules, thus affecting their spectral properties. Although the PF lifetime increase seems to be only subtle, it can be easily imaged due to a good signal-to-noise ratio.

Fig. 3

Mouse with a subcutaneous tumor after ${\mathrm{AlPcS}}_4$ injection. First row: before irradiation, second row: after $100\mathrm{J}/{\mathrm{cm}}^2$ irradiation. The third column displays PF kinetics from tumor, skin, and abdominal area as marked by rectangles in the PF image.

By studying a group of 10 mice, it was found that the enhancement of PF amplitude by PDT treatment of the tumor was quite variable, but mostly within the range of 20% to 80%. In contrary, the PF lifetime was usually increased by 0.5 ns, from 4.7 to 5.2 ns. Radiant exposures below $3\mathrm{J}/{\mathrm{cm}}^2$ did not induce measurable changes in PF, whereas exposures of 10 to $20\mathrm{J}/{\mathrm{cm}}^2$ usually caused significant changes in PF lifetime and intensity. Further changes in PF after delivering more than $30\mathrm{J}/{\mathrm{cm}}^2$ were small.

3.3.

Singlet Oxygen Feedback Delayed Fluorescence in a Live and Dead Mouse

Further, we examined basic properties of DF emission in vivo, and most importantly, we observed a dramatic loss of DF signal when the animal was sacrificed, thus supporting the hypothesis that the DF emission is indeed oxygen-mediated. These results are shown in Fig. 4. Interestingly, the DF signal from the abdominal area was comparatively much weaker than the PF signal, whereas DF from the tumor was very bright [Fig. 4(a)]. Therefore, PF and DF images provide different information. A probable explanation is that SOFDF, as a second-order process, is sensitive to the “local” concentration rather than the “average” concentration, and local microenvironments in tumor and abdominal area are different, as indicated by the different PF lifetimes (Fig. 3). DF time-resolved sequences with a gate width of $3\mu \mathrm{s}$ and delay increasing from 1 to $16\mu \mathrm{s}$ were captured, and Fig. 4(a) displays the first and fifth frames for illustration. Based on this sequence, DF lifetime image was calculated, providing DF lifetime of $\sim 8\mu \mathrm{s}$ in the tumor and slightly longer lifetimes ($\sim 9\mu \mathrm{s}$) in some other tissues. The measured DF lifetimes differed slightly in different animals ($n=5$) within the range of 8 to $11\mu \mathrm{s}$.

Fig. 4

PF and DF images from a mouse with tumor. (a) Live mouse, the DF lifetime image was calculated according to Eq. (3). (b) Dead mouse 10 min after the sacrifice, (c) a control mouse without ${\mathrm{AlPcS}}_4$ injection, (d) DF kinetics from different parts of the body, and (e) comparison of DF signal intensities in different parts of the body for live, dead, and control mice.

Figure 4(b) displays the PF and DF emissions from the same mouse 10 min after it was sacrificed, which caused rapid depletion of intracellular oxygen in tissues. While the PF changed only little, the DF emission from the tumor dropped dramatically, more than $6\times$. This experiment clearly proves that the DF emission is produced by the singlet oxygen feedback mechanism and not by other DF mechanisms, which on the contrary are enhanced in the absence of oxygen.²⁵ The experiment with a live/dead mouse was done with three different animals and provided consistent results.

Finally, Fig. 4(c) displays the signal from a live mouse that was not administered ${\mathrm{AlPcS}}_4$. This shows that the majority of the weak signal from the dead animal (previously injected with ${\mathrm{AlPcS}}_4$) was probably due to autofluorescence of the tissue and not related to ${\mathrm{AlPcS}}_4$ emission. Lifetimes of these residual DF signals were 13 to $17\mu \mathrm{s}$.

3.4.

Photodynamic Therapy Effect and Visualization of Lysosome Permeabilization

Subsequently, the effect of PDT and lysosome permeabilization on the parameters of DF was investigated. Tumors of mice were exposed to increasing light doses delivered by a 635-nm laser at 40 or $200\mathrm{mW}/{\mathrm{cm}}^2$ (with a smaller diameter beam).

In contrary to PF, larger light doses and lysosome permeabilization led to fading of DF [Fig. 5(a)]. After delivery of a larger light dose, the loss of DF proceeds also in dark without further irradiation (data not shown). The DF from the tumor usually faded out and gradually converged to the intensity of the signal from the surroundings, whereas the DF from the skin and surrounding tissues often increased a little, which could be due to the small light dose of scattered light that it receives, as shown later. Even though the DF signal drops and the tumor contrast decreases, there is still plenty of DF emission (high above the background levels), indicating that the generated ${}^1{\mathrm{O}}_2$ can still be detected. The PF emission from the tumor does not fade out even after prolonged periods of irradiation due to the excellent photostability of ${\mathrm{AlPcS}}_4$. The drop of the SOFDF signal after the lysosomal release could be explained by the fact that SOFDF is a second-order process in the concentration of triplet states, and as such it is weakened after dilution into cytoplasm. In addition, it is interesting to note that the DF lifetime is slightly shortened during the PDT, which is unexpected, because lengthening of lifetime due to oxygen depletion is rather anticipated.²¹

Fig. 5

PF and DF intensity and lifetime images from a mouse during PDT irradiation of the tumor with different radiant exposures. (a) The effect of 50 and $100\mathrm{J}/{\mathrm{cm}}^2$ on DF and PF intensities and lifetimes. (b) Small radiant exposure of $6\mathrm{J}/{\mathrm{cm}}^2$ caused an enhancement of DF while PF lifetime did not show marks of lysosome permeabilization. (c) Tumor was exposed by removing the skin, and the surroundings of the tumor were covered with a black tape to prevent any potential light scattering from outside of the tumor. (d) Irradiation of the tumor from the bottom side with $8\mathrm{J}/{\mathrm{cm}}^2$ and then from the top side.

Figure 5(b) shows that a relatively small radiant exposure ($6\mathrm{J}/{\mathrm{cm}}^2$), can lead, on the contrary, to a remarkable increase in DF signal (by 30% to 100% in different animals, $n=3$). In this specific case, this exposure did not induce an appreciable lysosome permeabilization in the tumor, as can be seen from the absence of a lifetime increase in the PF lifetime image.

It was verified that the time-resolved DF and PF signals could be imaged also from an exposed tumor after removal of the skin, as shown in Fig. 5(c), giving results consistent with skin-overlaid tumors. The surroundings of the tumor were covered with black material to prevent any potential scattering from tissue outside of the tumor.

Figure 5(d) shows PF and DF from a tumor that was irradiated first from the bottom side and then from the top side, thus simulating an inhomogeneous illumination, which is a common problem in PDT due to scattering and varying depth of the tumor. This illustrates the power of imaging over the point detection because different parts of the tumor are in different stages of PDT, as can be seen in the PF and DF images.

The experiments shown in Figs. 5(a)–5(d) were performed with groups of mice counting five, three, three, and three animals, respectively, and provided qualitatively consistent results. The experiment in Figs. 5(d) was done with different camera settings and therefore the absolute values of signal intensity shown in the images are not directly comparable to the images in remaining panels. Measurements of PF and DF intensities in a group of five mice showed that luminescence intensities from tumors in different individuals varied up to a factor of 2. Nevertheless, the qualitative observations described above were consistent across different animals.

3.5.

Singlet Oxygen Image

An image of SOFDF from a mouse with tumor was presented in Fig. 4(a). It is interesting to investigate what is the relation of such SOFDF image to the ${}^1{\mathrm{O}}_2$ image. SOFDF emission comes from the encounter of the excited triplet state with ${}^1{\mathrm{O}}_2$ and therefore the SOFDF emission rate is proportional to the product of the ${}^1{\mathrm{O}}_2$ concentration and triplet state concentration, as shown in Eq. (1). This implies that

The initial concentration of triplets generated after a laser pulse is usually proportional to the PF intensity, because we assume that the intersystem crossing rate and the radiative fluorescence rate are in a fixed proportion. Then the ${}^1{\mathrm{O}}_2$ image could be approximated by taking the SOFDF image and dividing it pixel by pixel with the PF image. This approach is validated in Fig. 6(a). A series of solutions of ${\mathrm{AlPcS}}_4$ in PBS in cuvettes was prepared, with concentrations gradually increasing from 0 to $7.5\mu \mathrm{M}$. Absorbances of all these samples were below 0.1 to avoid an inner filter effect and other artifacts. Under such conditions, the amount of generated triplet states scales linearly with the ${\mathrm{AlPcS}}_4$ concentration, and so does the amount of generated ${}^1{\mathrm{O}}_2$. (This is because the amount of ${}^3{\mathrm{O}}_2$ is much larger than the amount of triplet excited states, and a vast majority of triplets is quenched by oxygen to give rise to ${}^1{\mathrm{O}}_2$, thus making the amount of the generated ${}^1{\mathrm{O}}_2$ proportional and very similar to the amount of triplet states.) We can see in the graph in Fig. 6(a) that the ratio of intensities $I_{\mathrm{SOFDF}}/I_{\mathrm{PF}}$ is indeed linearly proportional to the amount of the generated ${}^1{\mathrm{O}}_2$.

Figure 6(b) then shows the ${}^1{\mathrm{O}}_2$-based image of a mouse reconstructed by dividing the DF image with a PF image pixel by pixel [background level measured in Fig. 4(c) was subtracted from the DF image first]. However, caution must be exerted when interpreting such an image. As shown earlier, the PF lifetime was different in the abdominal area, which indicated a different local microenvironment in the surroundings of ${\mathrm{AlPcS}}_4$ and likely also a different local concentration of excited states. Therefore, the SOFDF from the abdominal area was not comparable to SOFDF from the tumor and other parts of the body. This is indicated by the dashed line in Fig. 6(b), which approximately delimits the abdominal area.

Fig. 6

(a) ${\mathrm{AlPcS}}_4$ in PBS in cuvettes at concentrations ranging from 0 to $7.5\mu \mathrm{M}$. The graph shows the dependence of the average PF, DF, and DF/PF signal on the ${\mathrm{AlPcS}}_4$ concentration and amount of the generated ${}^1{\mathrm{O}}_2$. (b) ${}^1{\mathrm{O}}_2$ image approximated as the ratio of DF and PF. The plotted values of DF/PF ratio do not correspond to the absolute ratio of the emission rates, because the DF and PF images were captured with different camera settings (gain etc.). The dashed line indicates that the SOFDF intensity from the abdominal area cannot be easily compared to the SOFDF intensity in the rest of the body due to the very different microenvironment.

4.1.

Prompt Fluorescence Lifetimes

The following values of PF lifetimes in vivo were found: 5.2 ns in abdominal area, 4.7 ns in other parts of the body before irradiation, and 5.2 ns after lysosome permeabilization. PF lifetime of 6.2 ns was measured in a PBS solution at pH 7 (data not shown). Although there may be a systematic bias up to $\sim 0.5\mathrm{ns}$ due to the relatively broad laser pulse (3 ns) and gate width (3 ns), these values are in line with the previously reported PF lifetimes in solutions (5 to 6 ns) and cells (4 to 5 ns).^42,43,48 PF lifetime in acidic pH 4.1 was reported to be shorter by 0.5 ns, when compared to neutral pH 7.⁴⁸ This is one of the possible reasons for the PF lifetime increase after lysosomal permeabilization, because the interior of lysosomes is known to be more acidic than cytosol. It was also reported that aggregates of ${\mathrm{AlPcS}}_4$, which may be favored at high local concentrations inside lysosomes, have a short PF lifetime ($<1\mathrm{ns}$),⁴² but the time resolution of our time-gating (3 ns) does not allow to evaluate reliably such a fast luminescence component. While the increase of PF intensity certainly reports on lysosome permeabilization, it has a rather large variability, and time-gated lifetime imaging could be quite a robust technique to complement intensity measurements.

4.2.

Intensity of Singlet Oxygen Feedback Delayed Fluorescence Signal

The SOFDF-based ${}^1{\mathrm{O}}_2$ image in Fig. 6(b) was obtained with capture time of 6 s at 10 Hz laser repetition rate, and total radiant exposure of $20\mathrm{mJ}/{\mathrm{cm}}^2$, which is 3 orders of magnitude below the therapeutic radiant exposures. This illustrates the feasibility of real-time imaging during PDT. Moreover, the capture time and radiant exposure could be further significantly decreased by shifting the excitation from 690 nm to a shorter wavelength (630 to 660 nm). This would increase the absorption and would enable us to capture the main fluorescence peak around 680 nm, thus increasing the DF signal at least by a factor of 3 [Fig. 1(d)]. Excitation at 690 nm was used here because it is the minimal achievable wavelength with our tunable laser that overlaps with ${\mathrm{AlPcS}}_4$ absorption spectrum.

By comparing the intensity of PF and DF after the laser pulse, it was found that the integral DF from the tumor was 2000 to $6000\times$ weaker than the integral PF. With PF quantum yield of $\sim 0.4$,^42,43 we get a DF quantum yield of approximately ${10}^{-4}$ in the conditions of our experiment (it has to be remembered that the quantum yield of SOFDF, a second-order process, is a function of excitation intensity and photosensitizer concentration). Although this may seem a small number, the signal can be readily detected by an intensified gated camera, and it is about 3 orders of magnitude larger than the quantum yield of ${}^1{\mathrm{O}}_2$ luminescence.⁵ Moreover, the detection in the visible spectral region is technically less demanding than in the near-infrared.

4.3.

Comparison with Results Obtained in Cell Cultures In Vitro

When comparing our earlier-reported results in monolayer of cells in vitro²² with those obtained here in mouse in vivo, several similarities and several differences can be noticed. In both cases, lysosome permeabilization after irradiation usually caused an increase in PF intensity and a decrease in DF intensity. However, the decrease in DF was much more dramatic in vitro, whereas in vivo the DF signal is still high above the background levels even after prolonged irradiation. Moreover, a substantial initial increase in DF intensity was observed in vivo after a small radiant exposure, whereas this effect was much weaker or nonexistent in vitro. The ${\mathrm{AlPcS}}_4$ DF decay times in vitro were found to grow slightly after lysosome permeabilization (from 3.7 to $4.1\mu \mathrm{s}$), and more so when using ionic porphyrins.²¹ In contrary, in vivo, the DF of ${\mathrm{AlPcS}}_4$ decayed with lifetimes 7 to $11\mu \mathrm{s}$ and the lifetime slightly decreased during irradiation. This is rather unexpected because triplet lifetime is supposed to increase due to oxygen depletion during irradiation. In solutions, SOFDF lifetimes were found to report on oxygen concentration in the sample.^20,26 It would be very valuable to be able to infer such information also in vivo, but unfortunately the interpretation of the SOFDF decay times in this work is unclear. It is possible that in this specific case, the quenching by oxygen is not the most important triplet decay pathway.

4.4.

Interpretation of Singlet Oxygen Feedback Delayed Fluorescence Signal: Benefits and Limitations

We showed that the detected delayed emission can be indeed attributed to SOFDF. In tissue phantoms, this was shown by (1) quenching of DF by ${\mathrm{NaN}}_3$, (2) absence of DF in absence of oxygen, (3) enhancement of DF amplitude at higher oxygen concentration, and (4) rise–decay shape of the kinetics. In vivo in mouse, this was confirmed by a dramatic loss of DF signal after oxygen depletion achieved by sacrificing the animal (other types of DF are known to be enhanced at lower oxygen instead). The short rise-time of the SOFDF in vivo did not allow us to resolve the rise of the SOFDF kinetics.

In Fig. 2, we can notice that the overall DF emission intensity in tissue phantoms is stronger for the air-saturated sample than for the oxygen-saturated one, even though in the latter case the total amount of the generated ${}^1{\mathrm{O}}_2$ must be equal or somewhat larger. This illustrates an important caveat of the SOFDF-based sensing of ${}^1{\mathrm{O}}_2$: when the lifetime of ${}^1{\mathrm{O}}_2$ gets comparable or even larger than the triplet lifetime, which happens typically at high concentration of ${}^3{\mathrm{O}}_2$, the ${}^1{\mathrm{O}}_2$ population can “outlive” the triplet state population, and finding a collision partner for SOFDF becomes improbable. This typically leads to lowering of the integral SOFDF emission at high ${}^3{\mathrm{O}}_2$ concentrations, as discussed elsewhere.²⁵ However, this scenario is not common in vivo, where the oxygen concentration is significantly lower and hence the triplet lifetime is longer than in air-saturated solutions (many microseconds),⁴⁹ and ${}^1{\mathrm{O}}_2$ lifetime is shorter, below $1\mu \mathrm{s}$.²⁹ Under such conditions, SOFDF integral intensity is an increasing function of the amount of the generated ${}^1{\mathrm{O}}_2$.

Another caveat was illustrated in Fig. 6(b). It was suggested that the ${}^1{\mathrm{O}}_2$ image can be best approximated by dividing the SOFDF image with the PF image to account for different average concentrations of the photosensitizer in different regions. However, it was noted that the intensity of the SOFDF/PF image still cannot be easily compared between two regions with a very different local surroundings of the ${\mathrm{AlPcS}}_4$ molecules. To explain this more clearly, let us assume that the average concentrations of triplets and also the amount of generated ${}^1{\mathrm{O}}_2$ in two regions of the body are the same, but in one region the triplets are localized in lysosomes at very high local concentrations, whereas in the other region triplets are spread in the cytoplasm at a much smaller local concentration. The SOFDF signal will be much larger in the first region, because the probability of interaction between ${}^1{\mathrm{O}}_2$ and the triplet is higher, whereas in the second region the ${}^1{\mathrm{O}}_2$ is much more likely to be deactivated by concurrent chemical and physical quenching instead. As a result, SOFDF signals differ, even though the amounts of triplets and ${}^1{\mathrm{O}}_2$ were the same. For the same reason, the SOFDF signal drop after lysosome permeabilization does not necessarily mean that a smaller amount of ${}^1{\mathrm{O}}_2$ is produced, because the local concentration of triplets is dramatically changed and thus the comparison of SOFDF before and after the permeabilization is troublesome. Therefore, it must be always remembered that SOFDF rate is proportional to the product of the triplet and ${}^1{\mathrm{O}}_2$ local concentrations and not average concentrations. As such, SOFDF reports only on the local concentration of ${}^1{\mathrm{O}}_2$ in the immediate vicinity of the photosensitizer molecule and not on the average concentration of ${}^1{\mathrm{O}}_2$ over the whole cell.

This is certainly an important shortcoming, but it must be noted that a very similar problem will be encountered with any indirect method relying on a fluorescence probe. The amount of signal from the fluorescence probe will be proportional to the number of interactions between ${}^1{\mathrm{O}}_2$ and the probe molecules, and therefore dependent on the inherently heterogeneous local microscopic concentration of the probe, which cannot be easily determined nor controlled. Moreover, the distribution of the probe in the individual cellular compartments will be different from the distribution of the photosensitizer, leading to very different probe responses depending on where in the cell the ${}^1{\mathrm{O}}_2$ is produced (the intracellular diffusion radius of ${}^1{\mathrm{O}}_2$ is only $\sim 100\mathrm{nm}$²⁹ and hence it has to be sensed locally in the site of its production). With SOFDF, the colocalization of the probe with the site of ${}^1{\mathrm{O}}_2$ production is not an issue. To conclude, the dependence of the signal response on the local microenvironment is always present in any indirect method, and SOFDF is not an exception. Only the direct detection of the ${}^1{\mathrm{O}}_2$ near-infrared phosphorescence emission will not suffer from these issues. However, it is interesting to note here that, even in the case of the steady-state direct detection, one has to be cautious, because the amount of phosphorescence increases with the lifetime and the radiative rate of ${}^1{\mathrm{O}}_2$, which both depend on the local microenvironment.²⁹ To account for this at least partly, time-resolved detection would have to be used. Although this can be achieved with a point detection,⁶ time-resolved imaging would be quite challenging, given the extremely low signal levels.

Comparison of benefits and limitations of SOFDF-based ${}^1{\mathrm{O}}_2$ detection and other methods has been discussed in detail in our earlier works.^21,25 Despite the caveats discussed here, we believe that SOFDF-based sensing of ${}^1{\mathrm{O}}_2$ could be a valuable tool, given that all other techniques suffer from serious problems. To our knowledge, ${}^1{\mathrm{O}}_2$-based images of tumors with a comparable radiant exposure, capture time, and quality have never been reported before. Although SOFDF cannot be applied to all photosensitizers, it was reported that a wide range of other photosensitizers are also capable of SOFDF emission (${\mathrm{TPPS}}_4$, TMPyP, porphycenes, eosin, rose Bengal, etc.),^20,26 including protoporphyrin IX,²⁷ which is currently the most frequently used photosensitizer in clinical PDT (it is formed endogenously in cells after administration of δ-aminolevulinic acid). Therefore, SOFDF-based singlet oxygen sensing could be potentially performed with a wide range of photosensitizers.