2.1.

Clinical Photodynamic Therapy

Written informed consent was obtained from all patients for all procedures. The patient treatment and measurement protocol were approved by the Roswell Park Cancer Institute Institutional Review Board. Thirteen patients with biopsy proven SCC ($<3\text{-}\mathrm{mm}$ thick) were measured using DOS. Each patient received the PS HPPH in dextrose 5% (D5W), $4.0\mathrm{mg}/{\mathrm{m}}^2$ infusion over 1 h. Treatment light ($140\mathrm{J}/{\mathrm{cm}}^2$, $150\mathrm{mW}/{\mathrm{cm}}^2$, and 665 nm) was delivered $\sim 24\mathrm{h}$ later in the OR while the patient was under monitored anesthesia.³⁸ The treatment beam was centered on the lesion with 1 to 2 cm margins, with the total area of light exposure not exceeding $25{\mathrm{cm}}^2$. Biopsies of tumor tissue for the STAT3 crosslinking analysis were taken before and after treatment from the center of the illuminated region, immediately following the optical measurements. A schematic of the treatment protocol and DOS measurements is shown in Fig. 1.

Fig. 1

Treatment plan for patients receiving PDT. The optical measurements were performed in the OR, 24 h after HPPH administration. Biopsies were taken after the optical measurements, both before and after treatment light illumination. Follow-up biopsies were used to assess response.

Three months after PDT, a 3-mm punch biopsy from the treatment field was acquired for pathological response verification. Patients were examined at 3- to 6-month intervals thereafter. The responses reported here were determined from the 3-month biopsy and/or subsequent clinical observation. The biopsies were interpreted by a head and neck pathologist and were graded as shown in Table 1.

Table 1

Response classification.

2.2.

Experimental Setup

The instrument used in the clinic combined three diffuse optical methods: diffuse reflectance spectroscopy (DRS), diffuse fluorescence spectroscopy (DFS), and diffuse correlation spectroscopy (DCS) and has been described previously.^20,39,40 Briefly, the DRS setup consisted of a tungsten-halogen broadband white light (HL-2000-FHSA, Ocean Optics, Dunedin, Florida) as the source and the master channel of a two-channel spectrometer (S2000, Ocean Optics, Dunedin, Florida) as the detector. While the wavelength range for the spectrometer was 300 to 1100 nm, only a subset (520 to 820 nm) of this range was used for DRS. The light was directed to the target with one source fiber, the diffusely reflected light was collected with one detector fiber, and the source detector (SD) separation was 1.6 mm.

For the DFS setup, a 410-nm laser diode (Power Technologies, Little Rock, Arkansas) with an output power of 4 mW was used as the excitation source and the more sensitive slave channel of the spectrometer was used as the detector. A 500-nm longpass filter (450 Saffron, GAM products, Los Angeles, California) was placed in-line to reject the excitation light while passing the autofluorescence (AF) and PS fluorescence with a SD separation of 0.8 mm. The subset of wavelengths used for DFS was 600 to 770 nm.

The DCS setup consisted of a long coherence length, 785 nm laser (CrystaLaser, Reno, Nevada), four photon counting detectors (Avalanche Photodiode, Perkin Elmer, Waltham, Massachusetts) and a custom-built autocorrelator board (Correlator.com). The source was directed to the tissue by a multimode source fiber ($400\text{-}\mu \mathrm{m}$ core diameter) and collected with single-mode detector fibers ($5\text{-}\mu \mathrm{m}$ core diameter). The outputs from the photodetectors were sent to the correlator board to determine the intensity autocorrelation function and photon arrival times, which were then saved by the computer. The SD separations were 0.6, 1, 1.2, and 1.6 mm, providing a measure of blood flow at various depths in the tissue.

A custom stainless steel probe (5-mm diameter) was constructed to keep all fibers in place and for ease of sterilization. All three setups were mounted on a portable cart for facilitating transport between the laboratory and the OR. Noninvasive optical measurements were performed in the OR before and after PDT while the patient was under anesthesia. Measurements consisted of the surgeon placing the probe in soft contact with the tissue surface. At least five measurements were acquired from different points on the lesion. For each probe placement, measurements were acquired sequentially with DCS first followed by the spectral measurements.

2.3.

Optical Data Analysis

Analysis of our three DOS techniques has been discussed previously.^20,39,40 Blood flow was measured using DCS. Briefly, when photons scatter from moving blood cells, the intensity of collected diffuse light fluctuates with time. The normalized electric field is extracted from the autocorrelation function and the information about the motions in deep tissue can be extracted since the electric field autocorrelation function satisfies a diffusion equation. ⁴¹ An analytical solution is available for the Brownian diffusion model in reflectance measurements. The mean-square displacement ($⟨\mathrm{\Delta }{r}^2(\tau )⟩$) of the red-blood cells can be modeled as $⟨\mathrm{\Delta }{r}^2(\tau )⟩=6D_{\mathrm{B}}\tau$, for the case of diffusive motion; where $D_{\mathrm{B}}$ is the effective diffusion coefficient of the tissue scatterers. Empirically it has been observed that the diffusion model fits the autocorrelation curves and $\alpha D_{\mathrm{B}}$ characterizes the blood flow in deep tissue.^42–46 Here, $\alpha$ represents the probability that a scattering event in tissue emanates from a moving scatterer ($\alpha$ is generally proportional to tissue BVf). BFI is reported as the ratio of the measured blood flow parameter of tissue $\alpha D_{\text{Btissue}}$ to the measured blood flow parameter of a tissue-simulating phantom made by diluting 20% Intralipid (Fresenius Kabi, Grand Island, New York) in water ($D_{\text{Bintralipid}}$).

Reflectance data were analyzed using an empirical look-up-table.^40,47 Tissue absorption was expressed in terms of BVf, ${\mathrm{StO}}_2$, water fraction (fixed at 0.65), and HPPH content (cHPPH). The reduced scattering coefficient was modeled as Mie scattering.^48,49 A probe-specific look-up-table was constructed by measuring 64 phantoms with known volumes of added India Ink, Intralipid, and water.⁴⁷ The look-up-table incorporated probe design, construction, and instrument response into the light transport model.^50–52 To extract the optical properties and physiological parameters of interest, the trust-region-reflective nonlinear fitting algorithm (lsqnonlin, MATLAB^®) was used.^20,39 The look-up-table approach was tested with tissue-simulating phantoms with known values for scattering, BVf, and cHPPH. The average error for scattering quantification was 5.8%, average error for BVf was 6.8%, and the average error for cHPPH was 8.1%. The accuracy of extracting the ${\mathrm{StO}}_2$ was not tested with this approach.

Tissue fluorescence ($F_{\text{tissue}}$) was modeled as a linear combination of fluorophores including HPPH (fHPPH) and AF, and the fluorescence data from each patient measurement were fit to the model to extract amplitudes of AF and HPPH fluorescence.^20,39 To account for the effects of tissue optical properties on the measured fluorescence, a Monte Carlo-based correction factor was determined using the optical properties at the excitation and emission wavelength extracted with DRS.^53–55

2.4.

Statistical Analysis

For binary classification, the four grades shown in Table 1 were combined into two groups: positive responders (CR) or negative responders (PR, SD, and PD). Receiver operating characteristic (ROC) analysis was used to assess the strength of each measured parameter to predict a particular outcome. The threshold for the sensitivity and specificity of each predictor is the maximum of the Youden index. The area under the curve (AUC) was used to assess the overall performance of each classifier. In addition, a multiple regression method was used to combine two parameters into a single classifier for ROC analysis, which was performed using MATLAB^® and MedCalc software packages.

2.5.

Signal Transducer and Activator of Transcription 3 Crosslinking

This assay quantifies the cumulative PDT-induced photoreaction. When cells or tissue are treated with PDT, immediate damages occur through ROS generated by the light-activated PS (photoreaction). ROS oxidize, among others, lipids and proteins. A preferential target for oxidation is the covalent crosslinking of latent STAT3 proportional to the absorbed PDT dose.^36,37,56 The level of STAT3 crosslinking in the tumor was determined immediately following light treatment. Biopsy tissue was extracted by sonication in RIPA buffer. Aliquots of cleared extracts, containing $20\mu \mathrm{g}$ proteins, were separated under denatured condition on 6% SDS polyacrylamide gels and analyzed by immunoblotting for the level of homodimeric STAT3 relative to total STAT3 as reported previously.³⁶

3.1.

Predicting Clinical Response

ROC analysis was performed for each parameter using the binary response ($\text{responders}=1$; $\text{nonresponders}=0$) as the classification variable. The sensitivity, specificity [given in percent (%)], and normalized (norm.) AUC for individual parameters are presented in Table 2.

Table 2

Performance of individual parameters in classifying responders versus nonresponders.

The two best individual predictors of response were the change in ${\mathrm{StO}}_2$ ($\mathrm{\Delta }{\mathrm{StO}}_2$), ($\text{sensitivity}=100\%$, $\text{specificity}=67\%$, and $\mathrm{AUC}=0.70$) and the change in cHPPH ($\mathrm{\Delta }\mathrm{cHPPH}$) ($\text{sensitivity}=60\%$, $\text{specificity}=100\%$, and $\mathrm{AUC}=0.80$). As shown in Fig. 2(a) $\mathrm{\Delta }{\mathrm{StO}}_2$ is considered fair ($\mathrm{AUC}\ge 0.70$) and $\mathrm{\Delta }\mathrm{cHPPH}$ [Fig. 2(b)] is considered good ($\mathrm{AUC}\ge 0.80$). As Table 2 summarizes, all other individual parameters were considered fair or poor.

Fig. 2

ROC curves for the best parameters for predicting response. (a) $\mathrm{\Delta }{\mathrm{StO}}_2$ shows fair classification ($\mathrm{AUC}=0.70$) and (b) $\mathrm{\Delta }\mathrm{cHPPH}$ shows good classification ($\mathrm{AUC}=0.80$). (c) The combined classifier ($\mathrm{\Delta }{\mathrm{StO}}_2$ and $\mathrm{\Delta }\mathrm{cHPPH}$) is considered excellent for predicting pathological response ($\mathrm{AUC}=1.0$).

As we and others have indicated previously, individual parameters may not be good enough for accurate discrimination of responders from nonresponders.^20,39,57 Thus, the combination of multiple parameters into a single classifier was investigated with ROC analysis. A multiple regression model was used to combine several parameters into a single classifier. For each combination of two parameters, both pretreatment values and changes in these parameters were tested. The analysis showed that the best pair of classifiers was [$\mathrm{\Delta }\mathrm{cHPPH}$ and $\mathrm{\Delta }{\mathrm{StO}}_2$]. This combined classifier perfectly predicted responders and nonresponders [Fig. 2(c), $\text{sensitivity}=100\%$, $\text{specificity}=100\%$, and $\mathrm{AUC}=1.0$].

3.2.

Diffuse Optical Spectroscopy Parameters Correlate with Signal Transducer and Activator of Transcription 3 Crosslinking

As indicated in our previous studies,^{20,23,36,37,39} the oxidative homodimeric crosslinking of STAT3 is a molecular indicator of local photoreaction. Thus, we investigated the relationship between the PDT-related parameters obtained by noninvasive DOS measurements and the relative level of STAT3 modification determined in an extract from a single biopsy taken immediately after light treatment. There was no correlation between BVf and STAT3 or between ${\mathrm{StO}}_2$ and STAT3. The PS-related parameters, cHPPH and fHPPH, showed some correlation with STAT3; $r^2=0.46$ [Fig. 3(a)] and $r^2=0.41$ [Fig. 3(b)], respectively. As shown in Fig. 3(c), the parameter with the highest correlation to STAT3 crosslinking was the BFI measured before PDT ($r^2=0.61$). PDT induced substantial ($>40\%$) changes in all DOS parameters. However, there were weak correlations between these changes and the STAT3 crosslinking. The only parameter whose change showed some correlation with STAT3 crosslinking was change in BFI ($r^2=0.42$), as shown in Fig. 3(d).

Fig. 3

Correlation of pre-PDT measurements with STAT3 crosslinking. (a) cHPPH measured with DRS, (b) fHPPH measured with DFS, (c) BFI measured with DCS, and (d) percent change in BFI. ROC analysis shows that (e) pre-PDT BFI as well as (f) $\mathrm{\Delta }\mathrm{BFI}$ are good parameters for classifying high and low STAT3 crosslinking. Error bars represent standard deviation of multiple ($n\ge 5$) measurements per patient.

For effective therapy, there needs to be PDT-induced photoreaction in the lesion.^{20,36,37,39,40} To investigate the ability of DOS measurements to predict whether a lesion had a photoreaction above a certain threshold, ROC analysis was performed. The classification variable was STAT3 crosslinking with a threshold set at 5% (median of all patients). The threshold simplified the analysis as a binary classification between high crosslinking (higher than 5%) and low crosslinking (lower than 5%). Figure 3(e) shows the ROC analysis for the pre-PDT BFI values, which indicated the best correlation with STAT3 (86% sensitivity, 100% specificity, and $\mathrm{AUC}=0.93$). Figure 3(f) shows the ROC analysis for $\mathrm{\Delta }\mathrm{BFI}$ which, of all the post-treatment changes, correlated best with STAT3 crosslinking (71% sensitivity, 100% specificity, and $\mathrm{AUC}=0.91$). These results indicated that optical measurements have the potential to predict the PDT-induced photoreaction. A summary of the ROC analysis for all parameters at discriminating between high and low STAT3 crosslinking is shown in Table 3.

Table 3

Performance of individual parameters in classifying high versus low STAT3 crosslinking.