Longitudinal Photoacoustic Imaging of the Pharmacodynamic Effect of Vascular Targeted Therapy on Tumors Free

The development of novel therapeutics for the treatment of cancer is an attritional process, with only approximately 5% of lead compounds that enter clinical trials becoming regulatory approved therapies (1). A better understanding of the effect of drug action on the specific microenvironment target would greatly increase the translational relevance of preclinical studies, and therefore increase drug approval, which would save time and resources (2, 3). The use of biomedical imaging techniques can help elucidate how novel agents interact with the tumor target (4), by providing the means to observe how tumors respond, and how efficacious the novel drug is on a case-by-case basis. If translated to patients, this could provide additional information beyond simple diagnosis of disease, such as an indication of the interaction between tumor and host microenvironment, which coupled with monitoring of therapeutic response, could provide a personalized medicine approach to therapy (5). One class of cancer therapeutics that would benefit greatly from visualization of the tumor microenvironment is that of vascular targeted agents (VTA). VTAs target tumor blood vessels with the intent of either destroying existing tumor vasculature (vascular-disrupting agents, VDA) or inhibiting any further growth of the tumor vascular network (angiogenesis inhibitors; AI; ref. 6). Several VDAs, including OXi4503 (Combretastatin A1-diphosphate) employed in the current study, have been used in clinical trials for treating a wide range of solid tumors, including non–small cell lung cancer, prostate, ovarian, and anaplastic thyroid tumors (7, 8). It would therefore be advantageous to observe the effect of VTA action on tumor vasculature in situ. Ex vivo analysis of tumor vasculature using histochemical methods is a powerful technique, but requires large cohorts of animals for significant results and cannot be used to study the same tumors in a longitudinal manner (9). Intravital microscopy is commonly used to provide high-resolution images of tumor growth and vasculature, but typically requires either contrast agents or labeled cells, can only visualize small tumors or partial volumes, and is either an invasive terminal procedure or requires the implantation of a surgical window to replace the highly scattering skin layer (10). Tumor vascularization can be assessed by ultrasound without surgical exposition; however, visualization of microvascular change is limited without the use of microbubble contrast agents (11). Likewise, MRI (12) and CT angiography (13) can provide images of the vasculature of whole tumors, but both require the use of contrast agents to obtain high-resolution images. In addition, high-resolution micro-CT (14) requires high X-ray doses limiting the number of repeat measurements in longitudinal studies. Another technique capable of evaluating tumor vascular function after VTA treatment is dynamic bioluminescent imaging (BLI; ref. 15); BLI is straightforward to implement and capable of high-throughput, but lacks the spatial resolution to visualize tumor vasculature, and requires transfection of cell lines used and administration of exogenous substrate.

Photoacoustic imaging (PAI) is an emerging noninvasive technique that offers the prospect of overcoming these limitations (16–19). It relies on the generation of ultrasound waves through the absorption of low-energy nonionizing laser pulses by light absorbing molecules, such as hemoglobin. By detecting the time of arrival of these waves at the tissue surface, an image of the internal tissue structure can be reconstructed. Because PA image contrast is based on optical absorption, and hemoglobin absorbs light strongly at visible and near-infrared wavelengths, the microvasculature can be visualized with excellent visibility without employing exogenous contrast. Moreover, because the contrast is encoded on to acoustic waves, which are scattered much less than photons, PAI avoids the depth and spatial resolution limitations of purely optical imaging techniques such as intravital microscopy: with PAI, cm scale penetration depths with scalable spatial resolution ranging from tens to hundreds of micrometers (depending on depth) are achievable. In addition, there is the potential to acquire functional information via the measurement of both blood oxygen saturation (18) and flow (20). These factors offer the prospect of high-resolution label-free three-dimensional (3D) imaging of whole subcutaneous tumor vascular morphology and function in an entirely noninvasive manner (21–25), thus making PAI a promising tool for longitudinal studies of VTA action. The assessment of response to VTA-based treatment in preclinical models has been demonstrated in several preliminary PAI studies; Bohndiek and colleagues have shown antiangiogenic therapy can be observed to normalize tumor vasculature and regress growth of early vessel formation (24), whereas early feasibility studies have shown that PAI can visualize the short-term response to VDA-based treatment at one time point and for a single dose (23, 25, 26), and multiple time points within 6 hours (27) and 24 hours after VDA treatment (28).

In the current study, we advance previously reported research on VDA-based therapy assessed by PAI by investigating the effects of the VDA OXi4503 at multiple time points over 16 days on two colorectal xenografts with differing vascular architecture and pathophysiology (29, 30). We also compare the vascular effects of a range of OXi4503 doses in the same tumor models. A key requirement was the ability to visualize the often subtle changes in tumor vascular morphology produced by the VDA, particularly in poorly vascularized tumors or with low-dose therapy. This was made possible by using a custom-designed high-resolution PA scanner based on all-optical detection which provides more detailed 3D images than conventional piezoelectric-based PA scanners previously used for tumor imaging studies. Using this system, we show how PAI provides a detailed assessment of the evolution of the VDA-induced vascular destruction–regrowth cycle, the relationship between baseline tumor vascular pathophysiology and treatment response, and the effect of dose on pharmacodynamic response.

An all-optical photoacoustic scanner based on a Fabry–Perot (FP) polymer film ultrasound sensor was used to image the mice in widefield tomography mode (19). A schematic of the system is shown in Fig. 1. Its operating principles have been described in detail previously (23, 31, 32). Briefly, it consists of a tunable optical parametric oscillator laser system (Quanta Ray Pro-270/premiScan, Newport Spectra Physics/GWU Lasertechnik—not shown) which generates 7 ns excitation laser pulses at a pulse repetition frequency (PRF) of 50 Hz. The output of the laser is coupled into an optical fiber that produces a divergent beam that forms a large diameter (∼2 cm) spot on the FP sensor. The beam is then transmitted through the sensor head and into the adjacent tissue where it is absorbed producing broadband acoustic waves at ultrasonic frequencies. These waves propagate back to the sensor where they are detected at different spatial points and used to reconstruct a 3D image.

The sensor itself is a multilayer thin-film structure comprising two dichroic mirrors separated by a 22 μm thick polymer spacer thus forming an FP etalon. The mirrors are designed to be highly reflective between 1,500 and 1,600 nm but transparent to wavelengths between 590 to 1,200 nm to permit transmission of the excitation laser wavelength (640 nm in this study) through the sensor head. The photoacoustic waves generated in the tissue arrive at the FP sensor where they modulate its optical thickness. This produces a corresponding time-varying change in reflectivity which is read-out using a 1,550 nm interrogation laser beam that is focused on to the surface of the FP sensor. Forming a 3D image requires recording the spatial distribution of the incident photoacoustic waves over the surface of the sensor. This is achieved by optically scanning the interrogation beam point-by-point over an area on the sensor using a galvanometer-based x–y scanner. At each scan point, a time-resolved photoacoustic waveform is acquired in response to a single laser pulse. In this study, the scan area was 14 mm x 14 mm, the step size was 100 μm, and the element size (defined to a first approximation as the FWHM diameter of the focused interrogation beam) was 60 μm. For each scan, approximately 20,000 waveforms, each containing 800 time samples with a temporal sampling interval of 10 ns, were acquired without signal averaging and using an incident laser fluence of 5 mJ/cm², which is below the maximum permitted exposure for human skin (33). All images were acquired using an excitation wavelength of 640 nm as this was found to provide an acceptable compromise between contrast and penetration depth. The sensor provides a broadband frequency response with a -3dB bandwidth of 39 MHz (31).

3D photoacoustic images were reconstructed from the acquired waveforms, using a time reversal algorithm (34) with a correction for acoustic attenuation in tissue implemented using a time variant filtering method (35). The image reconstruction algorithm was implemented using k-Wave, an open-source MATLAB toolbox developed at University College London for the time-domain simulation, and reconstruction of PA and ultrasound wave fields (www.k-wave.org; ref. 36). Before reconstruction, the detected raw photoacoustic signals, arranged in a grid of size 141 × 141 × 800, were interpolated onto a 3 times finer x–y grid of size 423 × 423 × 800. The sound speed used in the reconstruction was selected using an autofocus approach, based on a metric of image sharpness (37). An exponential function normalization with respect to depth was applied to the reconstructed image data set as a first-order correction for the effect of optical attenuation to aid visualization. All images are displayed as maximum intensity projections (MIP) of the reconstructed 3D image data sets. In a lateral x–y MIP of a PA image data set, subsurface anatomy can be obscured by the large PA signal generated in the skin due to its strong optical absorption. Unless otherwise stated therefore, the skin contribution was removed by computing all x–y MIPs over the depth range z = 1–6 mm, thereby excluding contrast for depths below 1 mm.

In order to quantify changes in the tumor signal produced by the VDA, the tumor was manually segmented from the reconstructed PA image. The segmentation was implemented by viewing the 3D image data set and applying a mask created by tracing a boundary around the widest cross-section of the tumor in the lateral view. The mean signal intensity of all voxels above the noise floor in the segmented image was then obtained, with the noise floor chosen to be the 10th percentile of signal intensities of voxels within the segmented region.

The system provides a 14 mm × 14 mm lateral field-of-view, a penetration depth of approximately 10 mm, and lateral and vertical resolutions in the range 50–150 μm, depending on x–y position and depth. The total acquisition time is approximately 7 minutes, limited by the 50 Hz PRF of the excitation laser. An illustration of the in vivo imaging performance of the system is provided in Fig. 1B. This shows an example image of a subcutaneous SW1222 tumor xenograft implanted in the mouse flank acquired using an excitation wavelength of 640 nm. The images are displayed as lateral (x–y) and vertical (x–z and y–z) MIPs of the reconstructed 3D image data set; the x–y MIP was computed for the depth range z = 0–8 mm, thus the top 1 mm was not removed in this specific example. An animated volume-rendered image is also available online (Supplementary Video 1) which further illustrates the 3D nature of the image data set. The tumor is clearly visualized in the center of the x–y MIP (as indicated by yellow arrows) with normal vasculature surrounding it. The central region of the tumor exhibits relatively uniform albeit somewhat granulated contrast. This is broadly consistent with previous studies (23) which have shown that this tumor type exhibits a dense, homogeneously distributed vasculature with subresolution vessel spacing; hence, in this example, the tumor core is represented by a region of spatially averaged contrast, rather than a network of individually resolvable tumor vessels. The vertical x–z and y–z MIPs also show the tumor core and the surrounding nontumor vessels and demonstrate that it is possible to visualize the full thickness of the tumor. The system can thus image an entire subcutaneous tumor and its host environment noninvasively with high resolution in 3D to a depth of up to 8 mm as required for this study.

All in vivo experiments were performed in accordance with UK “Animals Scientific Procedures act” (1986) and adhered to the Workman and colleagues' (38) “Guidelines for the welfare and use of animals in cancer research.” Female nu/nu CD1 mice between the ages of 8 and 10 weeks were acquired from Charles River Laboratories and housed in barrier conditions within individual ventilated cages with food and water ad libitum. Two human colorectal carcinoma cell lines SW1222 and LS174T (29, 30) were used to establish xenografts on account of their different vascular architectures (the former being well vascularized, whereas the latter has a poorer and more heterogeneous blood supply) enabling the influence of tumor blood supply on therapeutic response to be studied. Cell lines were tested negative for mycoplasma using in-house PCR. The cells were prepared through sterile tissue culture techniques to a concentration of 5 × 10⁶ per 100 μL of serum free media. A 100 μL bolus of cells was injected s.c. into the right flank of the animals, with subsequent tumor growth measured using digital callipers and tumor volume calculated using the ellipsoid formula (39):

Tumors were size-matched to maintain similar volumes across groups, with an overall range of diameters between 6 and 8 mm. For dosing, a stock concentration of OXi4503 was prepared in sterile saline and animals injected i.v. at a ratio of 10 mL/kg. Total drug administered was dependent on the stock concentration used, with the volume of i.v. dose kept at 10 mL/kg. The mice were scanned by placing them on the FP sensor head and inserting a drop of water between the skin and the sensor to provide acoustic coupling.

Two studies were undertaken; the first investigated the long-term action of the VDA on the vasculature of both tumor types over an extended period of 16 days, a time period sufficient to observe the expected tumor blood vessel destruction and regrowth cycle. From here on, this is termed the “vessel regrowth” study. The second study evaluated the response of the two different tumor pathologies to differing levels of drug dosage. From here on, this is referred to as the “Dose response” study. For the vessel regrowth studies, n = 4 SW1222 and LS174T tumors were imaged at baseline before intravenously administering a single 40 mg/kg dose of OXi4503. Subsequent imaging was performed at days 1, 2, 3, 6, 9, 13, and 16 after dose along with calliper measurements of tumor volume. For the dose response arm of the experiment, groups consisted of n ≥ 4 animals for both cell lines at concentrations of OXi4503 of 40, 10, and 1 mg/kg, with a saline sham control. These groups were imaged at baseline and day 2 after dose, as this was the expected point of greatest response to treatment based on previous studies using OXi4503 (40).

Statistical analysis of the data was performed using Matlab. All values are reported as mean ± SE. For the tumor regrowth study, the change in PA signal, over the duration of the study, in the SW1222 and LS174T mice (n = 4 each) was tested for statistical significance difference using a two tailed Mann–Whitney test. In addition, a Pearson correlation analysis was used to determine the correlation between change in PA signal and change in tumor volume measured with callipers. For the dose response study, we carried out a one-way ANOVA Ominibus test on the means of the four treatment groups, 40 mg/kg, 10 mg/kg, 1 mg/kg, and control. Further post hoc pairwise multiple comparison tests were used to examine differences between individual paired groups, using Fisher modified least significant difference (MLSD) test (Fisher–Hayter). For all statistical tests, the significance level was P < 0.1.

For the regrowth study, mice with SW1222 and LS174T tumors were dosed with 40 mg/kg OXi4503 and imaged longitudinally over 16 days to study the effect of the drug on the tumor signal. Figure 2 shows images (x–y and x–z MIPs) of a subcutaneous SW1222 tumor in the same mouse at different time points over this period. The images show a large central area of signal void at 24 hours after treatment, with a surviving rim of signal observed around the location of the tumor periphery. Over the 16-day time frame observed, this area of signal void gradually reduces as new vasculature repopulates the destroyed central core of the tumor, until it has almost disappeared by day 16. By this time, the contrast in this central area has returned to a level that is comparable with the predose image. Broadly similar behavior, albeit with some variation in the extent and time course of the tumor signal, is observed in a second mouse with an SW1222 tumor (Fig. 3), and in other SW1222 tumors imaged but not shown. Images acquired over the same time period of 2 different mice with LS174T tumors are shown in Figs. 4 and 5. The pretreatment images in both cases illustrate the more heterogeneous and sparsely distributed vasculature of the LS174T tumor compared with the SW1222 tumor with individual tumor vessels now visible. After dosing, the reduction in signal is smaller, and its spatial distribution less distinct, than that seen for the SW1222 tumors. In addition, the treated LS174T xenograft tumors shows vessels in a more irregular pattern within the tumor region, with some areas lacking signal and with a less distinct peripheral rim signal to the SW1222 tumors. The near-complete absence of signal in the central part of the SW1222 tumor is not so clearly observed in the LS174T tumors, with some small areas of contrast remaining throughout the entire 16 days of the study. Moreover, the extent to which signal returns to the tumor core and reestablishes its baseline spatial distribution over this period is significantly less pronounced than for the SW1222 tumor.

To assess the time-course of the vascular destruction and regrowth quantitatively, the images acquired for each mouse in the study were segmented to isolate the tumor region (see Materials and Methods). The mean voxel intensity of the segmented region was then estimated and plotted for each time point. This data is shown for both tumor types in Fig. 6A and B alongside corresponding caliper-based measurements of tumor volume (Fig. 6C and D). All 4 SW1222 mice showed a marked decrease in signal of up to 80% over the first 2 days, consistent with the extensive regions of signal void in the tumors in Figs. 2 and 3. This initial reduction is followed by either an increase (m1_SW and m4_SW) or a prolonged decrease (m2_SW and m3_SW) in PA signal (Fig. 6), before a period of quiescence is established at around day 6. At approximately 13 days the signal starts to increase due to vessel regrowth, again as clearly observed in Figs. 2 and 3. The corresponding SW1222 tumor volume measurements in Fig. 6C show an initial reduction in tumor size for all 4 mice, tracking the drop in PA signal. Subsequently, with the exception of m3_SW, there is no significant growth in tumor volume over the remaining duration of the study.

The PA signal data for the LS174T tumor (Fig. 6B) show broadly similar behavior to that of SW1222, although several differences are evident. First, the extent to which the signal drops in the initial vascular destruction phase over the first 2 days is less pronounced than for the SW1222 tumors. The exception is m3_LS which shows a large spike (possibly due to a blood pool/clot), that appeared 24 hours after treatment and then plateaus out. Second, the extent to which the signal increases in the vessel regrowth phase (>12 days) is less evident between mice; m1_LS and m3_LS appear to show a modest increase in PA signal that could be indicative of vessel regrowth, whereas m2_LS and m4_LS exhibit a prolonged quiescent phase which extends to the study endpoint at day 16. Differences between the two tumor types are most evident in the tumor volume measurements (Fig. 6C and D) and their accompanying PA signal. Although three of the four LS174T tumors showed a small reduction in size over the initial 4 days (Fig. 6D), they did not exhibit the prolonged arrested growth of the SW1222 tumors (Fig. 6C) which appears to track the PA signal in most cases. The LS174T tumor volume of each mouse increases significantly beyond day 6 until the study endpoint, whereas the PA vascular signal (Fig. 6B) remains either constant or exhibits only a small increase.

For a statistical comparison of the extent of signal reduction and return in the SW1222 and LS174T during the study, the average signal reduction over 16 days was calculated for each mouse. The mean signal reduction in the SW1222 mice (n = 4) was 52.7 ± 8.80, whereas the corresponding reduction in the LS174T mice (n = 4) was 32.1 ± 10.3. A two tailed Mann–Whitney test shows a statistically significant difference in the reductions in PA signal between the two groups (P = 0.057), consistent with the qualitative and quantitative differences in Fig. 6A and B. To test for a correlation between the PA signal (Fig. 6A and B) and the tumor volume (Fig. 6C and D), at each time point both parameters were averaged for the 4 mice in each group. Using a Pearson correlation analysis, there was a strong positive correlation found between the PA signal reduction and the tumor volume measured by callipers for the SW tumors (r = 0.744, P = 0.055). For the LS tumors, there was no correlation (r = −0.02, P = 0.9636).

The second arm of the study, the dose response to OXi4503, was undertaken by comparing images acquired for doses of 40 mg/kg with 10 mg/kg, 1 mg/kg, and a saline sham control, with the results shown in Fig. 7. This figure shows x–y MIP images of all dose levels after dosing with OXi4503. The left-hand panel shows representative SW1222 tumors, whereas the right-hand panel shows LS174T tumors. Both panels show baseline and 48 hours after dose within the same tumor, with the tumor location indicated by the yellow arrows in the predose images. In SW1222 tumors, doses of 40 and 10 mg/kg produced the characteristic central tumor signal void, whereas the dose of 1 mg/kg produced a small but still visually discernible reduction in signal compared with sham-treated controls. In LS174T tumors, only the 40 and 10 mg/kg doses produce a notable effect, with the 1 mg/kg dose being comparable with the saline control in which no visually discernible effect was observed. To assess the dose response quantitatively, the tumor regions on the images for each mouse were segmented as described previously. The percentage reduction in mean voxel intensity from the corresponding predose baseline image was estimated and plotted for each dose and the control as shown in Fig. 7B. For the SW1222 tumors, mean reductions of 60%, 52%, and 20% of the medial PA signal intensity from baseline are observed for 40, 10, and 1 mg/kg groups, respectively. For the LS174T tumor, smaller reductions of 30%, 26%, and 4% are evident.

Using an ANOVA omnibus test, we found that for both the SW1222 and LS174T tumors, there was a statistically significant difference between the mean changes in PA signal from baseline to 48 hours of the dose groups. ANOVA (F(3,14) = 10.17, P = 0.0008) for SW1222 and ANOVA (F(3,11) = 4.35, P = 0.03) for LS174T. We then conducted post hoc pairwise multiple comparison tests to examine differences between individual paired groups, using Fisher MLSD test (Fisher–Hayter). For the SW1222 tumors, the changes in PA signal within the control group were significantly different from those within the 10 mg/kg group (P = 0.016) and 40 mg/kg group (P = 0.001). Similarly, PA signal changes in the 1 mg/kg group were also significantly different from the 10 mg/kg group (P = 0.029) and 40 mg/kg group (P = 0.0016). There were no significant differences between the control and 1 mg/kg groups, and also between the 10 and 40 mg/kg groups. For the LS174T tumors, statistically significant differences exist between the control group and the 10 mg/kg group (P = 0.043), and between the control group and the 40 mg/kg group (P = 0.046). There is no statistically significant difference between other dose group pairs.

In this study, we have shown that by undertaking serial longitudinal PA imaging studies we can observe the pharmacodynamic response of the tumor vasculature to therapy in a completely noninvasive manner over an extended period of time.

Both the qualitative and quantitative results shown are consistent with previous ex vivo histologic studies. First, consider the data in Figs. 2 to 6 which correspond to the vessel regrowth arm of the study. The initial vascular disruption produced by treatment is denoted by a lack of PA signal in the tumor core. This is clearly seen at the early stages (24 and 48 hours after dose) of response to OXi4503, and is significantly more pronounced in SW1222 tumors compared with LS174T. It can also be seen that the peripheral rim of the SW1222 tumors, most noticeably for mouse m4_SW (Fig. 2), shows a high signal intensity following treatment; this remaining viable rim following treatment by combretastatin compounds is well known through published histologic studies (41–43). The appearance of this rim is a sign of effective tumor-specific vasculature targeting and normal tissue sparing and is probably a consequence of several factors. The first factor is that vessels in the periphery of the tumor have survived through interaction with the surrounding normal tissue that is unaffected by the action of OXi4503, and these vessels have a high coverage of pericytes, promoting stabilization and making them less responsive to VDAs than the central, less mature vessels (44). The second is that any inflowing blood no longer has anywhere to go within the necrotic core of the tumor, and pools at the tumor edge, increasing the level of hemoglobin at the rim and therefore the PA signal. This has recently been demonstrated by optical projection tomography in both LS174T and SW1222 xenografts following OXi4503, which also revealed an increase in the size of peripheral vessels after treatment which could be an additional contributing factor to the increased PA rim signal (45). A third possible reason is the efflux of vascular and cellular debris including absorbing hemoglobin fragments from the destroyed tumor core being pushed to the periphery of the tumor due to interstitial pressure (46). This necrotic debris has previously been observed by transmission electron microscopy at 24 hours after treatment with OXi4503 (47).

The initial decrease in PA signal following administration of the VDA can be observed to diminish over time and is indicative of tumor revascularization. It is visually most noticeable in the SW1222 tumors, especially in Fig. 2 where the large central void created by the VDA appears to become almost completely repopulated by new vessels from the outside edge of the tumor mass over the 16 days of the study. The PA images also suggest some revascularization of the LS174T tumors occurs, though compared with the SW1222 tumors, this is attenuated, less well defined, and exhibits greater variability between mice, a consequence of the more heterogeneous vasculature exhibited by this tumor. Although the PAI data suggest a degree of revascularization in all tumors, Fig. 6A and B shows that no tumor of either SW1222 or LS174T returns fully to pretreatment levels of PA signal.

The difference in response of the SW1222 and LS174T tumors to the VDA is confirmed by a two-tailed Mann–Whitney test which shows a statistically significant difference in the average reduction in PA signal over the 16-day duration of the study between the two tumors (P = 0.057). The mean signal reduction in the SW1222 mice (n = 4) was 52.7 ± 8.80, whereas the corresponding reduction in the LS174T mice (n = 4) was 32.1 ± 10.3. It is well known that the two tumors have very different vascular states and pathophysiologies (29, 30, 45), with the well-oxygenated SW1222 tumors having a more homogeneous distribution of extensive, smaller vessels, whereas LS174T has a heterogeneous distribution with larger vessels toward the periphery and areas of poor vascularization, leading to extensive hypoxia. This makes the LS174T more inherently resistant to treatment with VDAs; not only are the larger vessels in the LS174T more mature and therefore more resistant to the drug, but LS174T cells are already more accustomed to hypoxic conditions (30) and thus better adapted to withstand the oxygen deficiency resulting from the VDA-induced vascular destruction. These differences in treatment response are reflected in the PA images. The more extensive and organized the original vascular network the greater the destruction observed, as seen in the PA images of the SW1222 tumors (e.g., Figs. 2 and 3); extensive areas of PA signal void correspond to avascular regions within these tumor types following treatment with OXi4503. By contrast, in the more chaotic and heterogeneous vascular architecture found in LS174T tumors (Figs. 4 and 5), there is a clear decrease in PA signal (Fig. 6B), but it is noticeably attenuated compared with the SW1222 tumors. The initial vascular state of the tumor also affects tumor growth. Figure 6A and C show that, in most cases, the drug-induced reduction in the SW1222 tumor PA signal is matched by an arrest in growth with tumor volumes typically remaining below baseline. A strong positive correlation was found between the reduction in PA signal and the measured tumor volume at each time point (r = 0.744, P = 0.055). By contrast, Fig. 6D suggests that LS174T tumor growth continues following an initial week of inhibition, despite the relative quiescence of the corresponding PA signal which suggests that the drug is continuing to inhibit vascular growth (Fig. 6B). Indeed, there was no correlation found between the PA signal reduction and tumor volume (r = −0.02, P = 0.9636).

Having shown that PAI can visualize major vascular disruption following treatment with a high dose of OXi4503, it is also notable that PAI can detect smaller changes in vascular disruption within the tumor, as demonstrated by reduced doses of the VDA. Figure 7 shows the effect of OXi4503 treatment at 40 mg/kg, 10 mg/kg, 1 mg/kg, and saline sham control at baseline and 48 hours for PA imaging in SW1222 and LS174T tumors. Again, the effect on LS174T tumors appears much less pronounced than in SW1222 tumors. In the case of the LS174T tumors, post hoc pairwise multiple comparison tests show a statistically significant difference between the vascular changes in the control group and the 10 mg/kg group (P = 0.043), and between the control group and the 40 mg/kg group (P = 0.046). There is no statistically significant difference between other group pairs. However, for the SW1222 tumors, there is a statistically significant difference between the control group and the 10 mg/kg group (P = 0.016), control group and 40 mg/kg group (P = 0.001), 1 mg/kg group and 10 mg/kg group (P = 0.029), and between the 1 mg/kg group and the 40 mg/kg group (P = 0.0016). These results are consistent with Hill and colleagues (35) who also found a dose-dependent response to OXi4503. We show here that measures of PAI can distinguish between effective (10 and 40 mg/kg) and noneffective (1 mg/kg) doses of the same vascular-disrupting drug, based on image assessment compared with baseline data.

In the current study, it has been assumed that the PA signal at 640 nm provides a measure of total hemoglobin. In practice, at this wavelength, the signal is also sensitive to changes in blood oxygen saturation (SO₂). This is a potentially confounding factor; for example, if the tumor SO₂ decreases over time, the PA signal at 640 nm will increase leading to an underestimation of the VDA-induced vascular destruction. However, the spatial–temporal distribution of the signal in both arms of the study is consistent with the known VDA response of the tumor models used (40, 48), suggesting that interpreting the PA signal as a marker primarily of tumor vascularization is reasonable in the context of the current study. Nevertheless, the potential for error due to SO₂ changes is recognized, particularly in terms of image quantification, the extent of which could be assessed in future by acquiring images at the SO₂-independent 808 nm isosbestic wavelength for comparison.

This study has demonstrated that the FP-based PA scanner can provide high-resolution images of deep (≈8 mm) subcutaneous tumors at a higher quality than conventional piezoelectric-based scanners. This is a consequence of the optical transduction mechanism it employs which provides a broadband frequency response, fine spatial sampling, and acoustically small element size all of which are important for an accurate image reconstruction. Additional factors contributing to the high image quality include the methods employed to mitigate the effects of acoustic attenuation, identifying the optimal sound speed and the use of an optimized reconstruction algorithm based on time-reversal. A current limitation however is the relatively long acquisition time (∼7 minutes), a consequence of the sequential manner in which the sensor is addressed and the relatively low PRF of the excitation laser. There is the potential to reduce this significantly and acquire 3D images in a few seconds by increasing the laser PRF, parallelizing the detection scheme (49), and employing compressed sensing techniques (50). Other anticipated technical advances include the development of methods for acquiring functional images of blood oxygen saturation using spectroscopic methods (18) and blood flow via the PA Doppler effect (20). Although both methods still require further development for use at multi-mm or cm scale depths relevant to imaging whole tumors, they offer the additional prospect of studying tumor hypoxia and the role it plays in VTA-based therapeutic response.

Although this study has focused on the use of PAI as a preclinical imaging tool, the technology and the results obtained in the current study are translatable to the clinic with a number of PA scanners currently being developed or evaluated for the clinical assessment of a variety of cancers including breast, prostate, and ovarian cancer (51). The visualization of vascular destruction and revascularization following treatment observed in the current study could be beneficial in clinical practice. For example, monitoring the time to regrowth could prove significant in the personalized clinical scheduling of repeat-dose treatment. In cases where the vessel regrowth originates from the surviving rim, specific targeting of the rim with combination therapies and using PAI to monitor the effect on the tumor could be advantageous in optimizing treatment parameters. The ability to observe differences in vascular pathophysiology, which affects outcome to treatment with VDAs, could potentially be used to assess initial tumor vascular state, and therefore expected outcome, in different patients. The observation of the pharmacodynamic effect of VDA action will allow personalized dose scheduling for subsequent treatments. Moreover, the use of functional PAI for the assessment of tumor hypoxia could yield a further predictor of response to specific therapies to help tailor, plan, and monitor treatment.

This study demonstrates the applicability of PAI for the evaluation of cancer therapy based on the use of VDAs. Using a custom-designed scanner, we have shown that high-resolution 3D images of the vascular morphology of whole subcutaneous tumors can be acquired longitudinally, noninvasively, and without use of exogenous contrast agents. The high quality of the acquired images and their 3D nature have been key to this endeavor, enabling the detection of subtle spatial–temporal vascular changes in response to the VDA. This capability has been exploited to determine the comparative initial effects of OXi4503 action on the different vascular architecture of two colorectal tumor models, and the subsequent patterns of vascular regrowth. The influence of their different vascular pathophysiologies on treatment response over 16 days, and the effect of different doses of OXi4503, is described. In doing so, we have demonstrated, for the first time in a longitudinal, noninvasive manner, that the tumor regrows inward from the surviving rim. PAI was also used to show that highly vascularized SW1222 tumors exhibit both greater initial vascular destruction in response to OXi4503 and a more pronounced subsequent long-term revascularization than the LS174T tumors, which possess a more sparsely distributed heterogeneous vasculature. In addition, we have shown that PA imaging is able to differentiate between the effects of different doses of OXi4503. Although this study has concentrated on tumor response to VDA treatment, other VTAs such as antiangiogenic therapies using humanized antibodies (52), and small molecules that attempt to normalize the tumor vascular microenvironment (53), would also be prime candidates for investigation with PAI.

In summary, we have demonstrated that PA imaging is a powerful preclinical tool for studying the tumor response to vascular-disrupting therapy. Moreover, PA imaging is well suited to clinical use, and the translational aspect of the current study offers new opportunities for personalized clinical medicine in terms of predicting the response to therapy and determining the most suitable dosing strategies for individual patients.

P. Beard holds ownership interest (including patents) in DeepColor SAS. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S.P. Johnson, O. Ogunlade, M.F. Lythgoe, R.B. Pedley

Development of methodology: O. Ogunlade, P. Beard, R.B. Pedley

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S.P. Johnson, O. Ogunlade, M.F. Lythgoe, R.B. Pedley

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.P. Johnson, O. Ogunlade, M.F. Lythgoe, P. Beard

Writing, review, and/or revision of the manuscript: S.P. Johnson, O. Ogunlade, M.F. Lythgoe, P. Beard, R.B. Pedley

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): O. Ogunlade

Study supervision: M.F. Lythgoe

The VDA OXi4503 (combretastatin A1-diphosphate/CA1P), currently in clinical trial for cancer treatment (NCT02576301), was a kind gift from Dr. David Chaplin (Mateon Therapeutics). This work was also supported by King's College London and University College London Comprehensive Cancer Imaging Centre, Cancer Research UK, and the Engineering and Physical Sciences Research Council (EPSRC), in association with the Medical Research Council and Department of Health, UK (C1519/A10331), and ERC Advanced Grant Ref: 741149.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.