Monitoring early response to chemoradiotherapy with ¹⁸F-FMISO dynamic PET in head and neck cancer

Abstract

Purpose

There is growing recognition that biologic features of the tumor microenvironment affect the response to cancer therapies and the outcome of cancer patients. In head and neck cancer (HNC) one such feature is hypoxia. Here we investigated the utility of ¹⁸F-fluoromisonidazole (FMISO) dynamic positron emission tomography (dPET) for monitoring the early microenvironmental response to chemoradiotherapy in HNC.

Experimental Design

72 HNC patients underwent FMISO dPET scans in a customized immobilization mask (0–30 min dynamic acquisition, followed by 10 min static acquisitions starting at ~95 min and ~160 min post-injection) at baseline and early into treatment where patients have already received 1 cycle of chemotherapy and anywhere from 5–10 fractions of 2 Gy per fraction radiation therapy. Voxelwise pharmacokinetic modeling was conducted using an irreversible one-plasma two-tissue compartment model to calculate surrogate biomarkers of tumor hypoxia (k₃ and Tumor-to-Blood Ratio (TBR)), perfusion (K₁) and FMISO distribution volume (DV). Additionally, Tumor-to-Muscle Ratios (TMR) were derived by visual inspection by an experienced nuclear medicine physician, with TMR>1.2 defining hypoxia.

Results

135 lesions in total were analyzed. TBR, k₃ and DV decreased on early response scans, while no significant change was observed for K₁. The k₃-TBR correlation decreased substantially from baseline scans (Pearson’s r = 0.72 and 0.76 for mean intratumor and pooled voxelwise values, respectively) to early response scans (Pearson’s r = 0.39 and 0.40, respectively). Both concordant and discordant examples of changes in intratumor k₃ and TBR were identified; the latter partially mediated by the change in DV. In 13 normoxic patients according to visual analysis (all however having lesions with TMR=1.2), subvolumes were identified where k₃ indicated the presence of hypoxia.

Conclusion

Pharmacokinetic modeling of FMISO dynamic PET reveals a more detailed characterization of the tumor microenvironment and assessment of response to chemoradiotherapy in HNC patients than a single static image does. In a clinical trial where absence of hypoxia in primary tumor and lymph nodes would lead to de-escalation of therapy, the observed disagreement between visual analysis and pharmacokinetic modeling results would have affected patient management in <20% cases. While simple static PET imaging is easily implemented for clinical trials, the clinical applicability of pharmacokinetic modeling remains to be investigated.

Introduction

There is growing interest in studying essential biologic features of cancer, such as metabolism, hypoxia, and expression of transmembrane receptors, noninvasively and repeatedly, during therapy. In parallel, efforts are underway to adapt therapies based on early treatment response reflecting cancer biology rather than simple structural changes (e.g., changes in tumor volume or shape). Such biologic changes may be a crucial component in the effective management of many cancer patients [1–5]. An improved understanding of biologic processes and their relationship to functional imaging data and tumor response may yield pertinent information that can be applied in a rational way such as adaptive radiotherapy [6], patient stratification in clinical trials (e.g., utilizing hypoxia-activated prodrugs; [7]), development of novel pharmaceutical drugs [8] or optimization of treatments utilizing combinations of targeted drugs with radiotherapy [9].

Hypoxia is a common feature of solid tumors and a major negative prognostic indicator regarding patient outcome after radiotherapy [10]. A meta-analysis of clinical trials in patients with head and neck cancer (HNC) undergoing radiotherapy with curative intent concluded that there is evidence in favor of targeting hypoxia through various treatment modifications [11], although discordant results have also been reported [12]. Hypoxic modification continues to have a limited impact on clinical practice, likely due to the non-selectivity of patients entered on clinical trials [13].

Positron emission tomography (PET) using hypoxia selective imaging probes has the ability to detect tumor hypoxia non-invasively. PET imaging with ¹⁸F-fluoromisonidazole (FMISO), the most extensively studied hypoxia PET radiotracer, has been shown to be clinically feasible [14, 15] and reproducible [16, 17]. However, Tumor-to-Blood Ratio (TBR) or Tumor-to-Muscle Ratio (TMR), metrics derived from static FMISO acquisitions that are commonly used as surrogate biomarkers of tumor hypoxia, depend on the exact time point at which images are obtained; in addition, TBR and TMR also depend on the FMISO distribution volume that is non-uniform both within and between tumors [18]. Therefore, presence and extent of hypoxia may be over- or underestimated with this approach. A method that allows for more accurate quantification of hypoxia may help to improve confidence in interpreting hypoxia PET imaging data, in particular if the presence of hypoxia, and changes in hypoxia under treatment, are to be used to guide patient management.

Pharmacokinetic modeling of FMISO dynamic PET (dPET) in principle enables the derivation of a surrogate biomarker that more directly reflects hypoxia-mediated entrapment, while simultaneously providing information on tumor perfusion that may be important in distinguishing different tumor subtypes [19]. Tumor perfusion is an important measure of drug delivery, and changes in perfusion may provide an insight into the response to either nonspecific or targeted (e.g. antiangiogenic) therapies [20]. We recently demonstrated that simultaneous, multiparametric and single modality imaging of tumor hypoxia and perfusion with FMISO dPET can be performed in HNC patients with acquisition protocols with overall scan times as low as 20 min, comparable in duration to diagnostic ¹⁸F-Fluorodeoxyglucose (FDG) PET [18].

Complex relationships between tumor hypoxia, perfusion and glucose metabolism support the use of composite metrics providing additional information on tumor biology that may help in designing adaptive therapies [8, 21] and may result in better prediction of patient survival [22, 23]. Of note, these characteristics of the tumor microenvironment may not change in a uniform manner under therapy [8].

In the present study, tumor hypoxia, perfusion and FMISO distribution volume were assessed simultaneously from FMISO dPET scans in 72 head and neck cancer patients at baseline and after 1 cycle of chemotherapy concurrent with anywhere from 5 to 10 fractions of 2 Gy radiation (early response). We hypothesized that pharmacokinetic modeling of FMISO dPET may enable a more comprehensive assessment of tumor response than can be obtained from single static images.

Materials and Methods

Patient selection

Patients with HNC were enrolled prospectively to Institutional Review Board approved multi-arm trial (Protocol #04-070Protocol, NCT00606294 on www.clinicaltrials.gov; inclusion criteria: 18+ age, histologically confirmed diagnosis of squamous cell carcinoma of the head and neck, Karnofsky performance status of ≥70; exclusion criteria: all nasopharyngeal, paranasal sinus, salivary cancer and thyroid malignancies, prior chemotherapy or radiotherapy within the last three years, previous surgical resection for the same disease, any prior radiotherapy to the head and neck region, and pregnancy). All patients signed a written informed consent regarding the examination and use of anonymous data for research and publication purposes. This article reports the results of a small sub-study of the currently ongoing trial (72 patients, aged 60±11 years (range, 25–86 years); Table 1). Flowchart of patient selection and inclusion is presented in Figure 1.

Table 1.

Subject demographics and clinical characteristics.

	N=72 (%)
Gender
Male	66 (92)
Female	6 (8)
Age at diagnosis, y
<40	1 (1)
40–49	10 (14)
50–59	30 (42)
60–69	20 (28)
70–79	6 (8)
80+	5 (7)
Clinical tumor classification
T1	16 (22)
T2	36 (50)
T3	10 (14)
T4	8 (11)
Unknown	2 (3)
Tumor subsite
Base of tongue	39 (54)
Tonsil	28 (39)
Hypopharynx	1 (1)
Supraglottic larynx	2 (3)
Unknown primary	2 (3)
Clinical lymph node classification
N1	7 (10)*
N2a	7 (10)
N2b	31 (43)
N2c	25 (35)
N3	2 ( 3)
HPV status
Positive	47 (85)
Negative	8 (15)
Testing not performed	17
p16 status
Positive	5 (8)
Negative	60 (92)
Testing not performed	7
Chemotherapy regimen
Carboplatin/5-fluorouracil	3 (4)*
Carboplatin/paclitaxel	3 (4)
Cetuximab/cisplatin/bevacizumab/	1 (1)
Cetuximab	7 (10)
Cisplatin	55 (76)

^*Numbers do not total 100% due to rounding.

Figure 1.

Flowchart of patient selection and inclusion.

*According to the visual analysis by nuclear medicine physician.

PET/CT image acquisition and analysis

All patients underwent baseline FDG PET/CT scans of the head and neck for radiotherapy planning purposes, followed after 8±6 days by a baseline FMISO dynamic PET/CT scans (FMISO1). A second FMISO scan (FMISO2) was acquired at 17±5 days after FMISO1, after patients have already received 1 cycle of chemotherapy and anywhere from 5–10 fractions of 2 Gy per fraction radiation therapy. For all scans, patients were positioned on a flat-top couch wearing a customized radiotherapy treatment immobilization mask that minimizes motion and allows for accurate repositioning. Intravenous lines inserted in antecubital veins were used for radiotracer injection. Patients were administered an intravenous bolus injection of 390±14 MBq of FMISO. Each FMISO dynamic PET acquisition consisted of 3 segments: (i) at time t=0, a 30 min acquisition binned into 6×5-sec, 3×10-sec, 4×60-sec, 2×150-sec, 2×300-sec and 1×600-sec frames, (ii) a 10 min static acquisition, starting at 93±6 min (FMISO1) and 96±9 min (FMISO2) and (iii) a 10 min static acquisition starting at 160±13 min (FMISO1) and 161±13 min (FMISO2) post-injection.

For both FMISO1 and FMISO2, all three FMISO segments were spatially co-registered using the rigid-body transformation calculated with General Co-Registration™ tool (General Electric Advantage Workstation v4.7) applied to their corresponding CT scans, and subsequently merged to form a concatenated FMISO dPET scan. These two scan series were also co-registered with the FDG PET scan. Lesions were delineated on the FDG PET images, using the adaptive threshold algorithm in the PET VCAR™ (Volume Computer Assisted Reading; General Electric Advantage Workstation v4.7).

Additionally, Tumor-to-Muscle Ratios (TMR) for both FMISO1 and FMISO2 were derived by visual inspection of the three FMISO segments by an experienced nuclear medicine physician (H.S.) as reported previously [6], independently of the medical physicist’s (M.G.) pharmacokinetic modeling. Spherical regions of interest were placed in tumors and metastatic lymph nodes, and activity concentration was measured using max SUV in the tumor volume of interest (VOI). Reference regions were placed in contralateral neck muscles showing low background activity and measured using mean SUV. Hypoxia-positive was defined as FMISO uptake in known sites of disease that was increasing or stable from scan time point i) to time point iii) of each scan series, with a TMR>1.2. Since TBR_max was defined similarly (max SUV in the tumor VOI divided by mean SUV in the jugular vein VOI), synchronization of delineation methods was not required.

Pharmacokinetic modeling

Voxelwise pharmacokinetic modeling of FMISO dPET images was performed in PMOD v3.604 (PMOD Technologies GmbH), utilizing an irreversible one-plasma two-tissue compartment model [18]. The output of the pharmacokinetic modeling are three kinetic rate constants: K₁, the kinetic rate constant describing transport from the vascular compartment to the extravascular tissue compartment and a surrogate biomarker of tumor perfusion, k₂, the kinetic rate constant describing the transport of FMISO from the extravascular tissue compartment back to blood and k₃, the kinetic rate constant approximating the rate of irreversible binding of FMISO and a surrogate for hypoxia-mediated entrapment. An important composite parameter DV (distribution volume) is given by

$$DV=v_B+\left(1-v_B\right)\frac{K_1}{k_2},$$ (1)

where v_B is the fractional vascular volume also calculated by the model. DV represents the total FMISO distribution volume i.e., overall concentration of unbound FMISO relative to blood, and sets the threshold above which FMISO accumulation is due to hypoxia specific binding. DV was previously shown to be non-uniform across a tumor upon tracer equilibration [18]. The input function (IF) was delineated on ipsilateral (with respect to the lesion) jugular vein on the early dynamic frame with the highest image intensity, by selecting ~100 hottest voxels. TBR was calculated from the 3^rd FMISO acquisition, using the measured activity concentration in the IF volume of interest, C_p(t) as the reference,

$$TBR=\frac{C_{target}(t)}{C_p(t)}.$$ (2)

The parameters v_B and DV are unitless, whereas K₁ (assuming unit density tissue), k₂ and k₃ are expressed in units of min⁻¹.

Pharmacokinetic modeling was also performed for ipsilateral (with respect to the primary lesion) sternocleidomastoid muscle, adipose tissue posterior to the cervical vertebrae and ipsilateral submandibular and parotid glands, on both baseline and early response scans.

Statistical analysis

All correlations were performed using Pearson’s correlation coefficient r (measure of correlation strength). Linear regression was performed to model the relationship between TBR (dependent variable) and k₃ (explanatory variable), and to calculate the slope coefficient (measure of correlation direction) and intercept coefficient. Comparisons of investigated metrics as calculated from (i) baseline and early response scans or (ii) primary lesions and lymph nodes were performed with unpaired 2-tailed Student t test. p<0.05 was assumed to represent statistical significance.

Results

In total, 135 lesions were analyzed (54 primaries and 81 lymph nodes) by both pharmacokinetic modeling and, independently, by visual analysis. The results of these analyses were subsequently compared side by side by nuclear medicine physician, nuclear medicine physicists and the radiation oncologists. The lesions ranged from 2 to 54 cm³ (median volume: 10 cm³). HPV (by in situ hybridization) and p16 testing (by immunohistochemistry) was performed in 56 and 64 patients, respectively, with 84% and 92% of tested patients harboring HPV positive and p16 positive tumors, respectively. 4 out of 72 patients were deceased at the time of writing, while additional 2 patients experienced local disease relapse (median follow-up period: 26 months, range: 5–73 months).

Mean values determined for all investigated parameters (i.e., TBR, k₃, DV, K₁ and v_B) as well as means from voxelwise analysis of pooled data are summarized in Table 2. Changes between early response and baseline FMISO scan data are displayed in Figure 2 for TBR, k₃, DV, K₁ and v_B. There is a visible shift to reduced values for both TBR and k₃, indicative of a decrease in hypoxia early during treatment. Baseline k₃ was the only metric that was significantly different between primaries and lymph nodes (Supplementary Table S1).

Table 2.

Summary of investigated metrics for the baseline and early response FMISO dynamic PET scans in tumors (n=135).

	Tumor-to-blood Ratio	k3 (min−1)	Distribution volume	K1 (min−1)	vB
Baseline (mean intratumor)	1.35±0.31	0.0029± 0.0022	0.97± 0.14	0.38± 0.17	0.20± 0.10
Early response (mean intratumor)	1.12±0.15	0.0020± 0.0016	0.91± 0.14	0.41± 0.21	0.21± 0.11
p-value*	<0.01	<0.01	<0.01	0.20	0.28
Mean difference	−0.22±0.26	−0.0009± 0.0021	−0.06± 0.14	0.03± 0.21	0.01± 0.12
Baseline (voxelwise)	1.48±0.49	0.0039± 0.0037	0.97± 0.18	0.35± 0.21	0.19± 0.12
Early response (voxelwise)	1.18±0.21	0.0021± 0.0022	0.92± 0.18	0.38± 0.25	0.22± 0.14

^*Unpaired 2-tailed Student t test.

Figure 2. Voxelwise distributions of investigated parameters derived from baseline and early response FMISO dynamic PET scans.

(A) Normalized distribution histogram of voxelwise TBR values at baseline (blue) and after 1 cycle of chemoradiotherapy (red), derived from pooled data.

(B) Equivalent histogram for k₃.

(C) Equivalent histogram for DV.

(D) Equivalent histogram for K₁.

(E) Equivalent histogram for v_B.

Corresponding values for normal tissues in the head and neck region are presented in Table 3. Apart from a small but significant decrease in TBR and DV values in submandibular gland, none of the investigated parameters were significantly different between baseline and early response scans for any of analyzed normal tissues.

Table 3.

Summary of investigated metrics for the baseline and early response FMISO dynamic PET scans in normal tissues (n=72 patients).

		Target-to-blood Ratio	k3 (min−1)	Distribution volume	K1 (min−1)	vB
Sternocleidomastoid muscle	Baseline	1.01±0.08	0.0007± 0.0006	0.89± 0.10	0.08± 0.04	0.05± 0.03
	Early response	1.00±0.09	0.0008± 0.0007	0.89± 0.12	0.09± 0.04	0.06± 0.04
	p-value*	0.46	0.78	0.98	0.24	0.13
	Mean difference	−0.01±0.09	0.0000± 0.0007	0.00± 0.07	0.01± 0.05	0.01± 0.04
Adipose tissue	Baseline	0.27±0.07	0.0004± 0.0004	0.24± 0.06	0.02± 0.01	0.01± 0.01
	Early response	0.25±0.07	0.0004± 0.0005	0.23± 0.06	0.02± 0.01 0.02 0.02	0.01± 0.01 0.01± 0.01
	p-value*	0.11	0.69	0.40	0.85	0.84
	Mean difference	−0.02±0.05	0.0000± 0.0006	−0.01± 0.05	0.00± 0.02	0.00± 0.01
Submandibular gland	Baseline	1.02±0.08	0.0002± 0.0003	1.04± 0.11	0.74± 0.53	0.16± 0.10
	Early response	1.00±0.09	0.0003± 0.0004	0.98±0.10	0.76± 0.40	0.18± 0.11
	p-value*	0.03	0.19	<0.01	0.86	0.30
	Mean difference	−0.03±0.08	0.0001± 0.0005	−0.06± 0.10	0.01± 0.55	0.02± 0.13
Parotid gland	Baseline	0.90±0.11	0.0002± 0.0004	0.88± 0.15	0.50± 0.32	0.10± 0.08
	Early response	0.87±0.11	0.0003± 0.0005	0.84± 0.11	0.56± 0.32	0.12± 0.09
	p-value*	0.13	0.17	0.09	0.35	0.37
	Mean difference	−0.03± 0.09	0.0001± 0.0006	−0.04± 0.10	0.05± 0.32	0.01± 0.10

^*Unpaired 2-tailed Student t test.

Pearson’s r between mean voxelwise k₃ and TBR in baseline scans was 0.72 (95% CI: 0.63–0.79; Fig. 3A). Each point in Figure 3A is color coded according to the mean DV for that tumor, illustrating that nonuniformity in the tumor DV reduces the k₃-TBR correlation over the patient population as a whole. These analyses were also performed on a voxel by voxel basis for the entire voxel data ensemble from all tumors. The results are displayed as an intensity histogram for the pooled voxelwise k₃-TBR data with the regression line (r = 0.76; Fig. 3B). Lower correlation between k₃ and TBR was observed in the results from early response scans, both on intratumor (r = 0.39, 95% CI: 0.24–0.53; Fig. 3C) and voxelwise (r = 0.40; Fig. 3D) levels. The change in mean intratumor k₃ (Δk₃ = k_3,FMISO2 - k_3,FMISO1) positively correlated with the change in mean intratumor TBR (ΔTBR = TBR_FMISO2 - TBR_FMISO1; r = 0.70, p<0.01; Supplementary Figure S1A). Additionally, Δk₃ negatively correlated with the change in mean intratumor DV (ΔDV = DV_FMISO2 - DV_FMISO1; r = −0.56, p<0.01; Supplementary Figure S1B). The correlations between metrics as calculated from baseline and early response scans were: r(TBR_FMISO1,TBR_FMISO2) = 0.58, r(k_3,FMISO1,k_3,FMISO2) = 0.41, r(DV_FMISO1,DV_FMISO2) = 0.53, r(K_1,FMISO1,K_1,FMISO2) = 0.41 and r(v_B,FMISO1,v_B,FMISO2) = 0.39 (p<0.01 in all cases). The correlations between metrics as calculated from baseline scans and their change after early response scans were r(TBR_FMISO1, ΔTBR) = −0.90, r(k_3,FMISO1, Δk₃) = −0.75, r(DV_FMISO1, ΔDV) = −0.53, r(K_1,FMISO1, ΔK₁) = −0.43 and r(v_B,FMISO1, Δv_B) = −0.48 (p<0.01 in all cases).

Figure 3. Correlation between k₃ and TBR as derived from baseline and early response FMISO dynamic PET scans.

(A) Scatterplot (one point per patient tumor) of the mean intratumor k₃ versus TBR at baseline, with linear regression fit (solid line). The points are color-coded according to the magnitude of the mean intratumor DV at baseline.

(B) k₃-TBR intensity histogram consisting of the pooled voxels from all tumors from baseline data, with the linear regression fit (solid line).

(C) Scatterplot of mean intratumor k₃ and TBR from early response scans, with included linear regression fit (dashed line). Superimposed is the linear regression fit from baseline data (A; solid line).

(D) Corresponding intensity histogram from early response scans, with included linear regression fit (dashed line). Superimposed is the linear regression fit from baseline data (B; solid line).

Pharmacokinetic modeling facilitated a more detailed early response assessment to chemoradiotherapy. This is best revealed by investigating individual patient cases that highlight the diversity of individual responses. The parametric maps for two patients reveal a concordance (Fig. 4A) and discordance (Fig. 4B) in changes in TBR and k₃, two surrogate metrics of tumor hypoxia derived from static and dynamic PET scans, respectively. For patient #1, the decrease in k₃ (Δk₃ = −0.0066 min⁻¹) was reflected in the decrease in TBR (ΔTBR = −1.26), with no discernible change in DV (Fig. 4A). However, for patient #2 (Fig. 4B), the decrease in TBR (ΔTBR = −0.17) and DV (ΔDV = −0.24) was observed whereas k₃ increased (Δk₃ = 0.0026 min⁻¹).

Figure 4. Example of concordant and discordant changes in k₃ and TBR.

(A) Patient #1 with 37 cm³ T4aN2c, HPV-positive, p16-positive tumor originating in the tonsil. Results from baseline and early response scans are presented in top and bottom rows, respectively. From left to right: Sagittal view of the late 10-min FMISO PET/CT scan, tumor-to-blood ratio (TBR) map of intratumor voxels, k₃ map representing hypoxia-mediated entrapment of FMISO and FMISO distribution volume (DV), representing overall concentration of unbound FMISO relative to blood. For this patient, both TBR and k₃ decreased after 1 cycle of chemoradiotherapy, while DV did not change substantially. ΔTBR = −1.26 (ΔTBR_max = −2.82), Δk₃ = −0.0066 min⁻¹ (Δk_3,max = −0.0160 min⁻¹), ΔDV =−0.01.

(B) Corresponding images for Patient #2 with 44 cm³ T1N2b, HPV-negative, p16-negative tumor originating in the supraglottic larynx. For this patient, TBR decreased, while k₃ increased. Decreased DV further contributed to underestimation of TBR. ΔTBR = −0.17 (ΔTBR_max = −0.71), Δk₃ = 0.0026 min⁻¹ (Δk_3,max = 0.0059 min⁻¹), ΔDV =−0.24.

TMR readings for both baseline and early response FMISO scans were available for 129 out of 135 lesions (for the remaining 6 lesions, TMR readings were not made on early response scans due to indiscernible FMISO uptake, thus assuming the resolution of hypoxia). Pearson’s r between TBR_max (derived as part of pharmacokinetic modeling) and TMR (derived independently by a nuclear medicine physician) was 0.94 and 0.84 on baseline and early response scans, respectively. However, Pearson’s r between k_3,max and TMR was 0.66 and 0.33 on baseline and early response scans, respectively.

On baseline scans, TMR was ≤1.2 in 22 lesions, suggesting absence of hypoxia, with 18 out of these 22 lesions having TMR≤1.2 also on early response scans (in the remaining 4 lesions, TMR increased to 1.3). TMR≤1.2 was observed in 56 lesions on early response scans (included are lesions for which FMISO uptake was indiscernible). In 21 out of these 56 lesions (8 primaries and 13 lymph nodes), 4+ consecutive voxels with k₃ ≥0.0026 min⁻¹ were identified on early response scans, suggesting the presence of some hypoxia by pharmacokinetic modeling. The k₃ threshold was derived as mean k₃ + 3 standard deviations of the value as measured in non-diseased sternocleidomastoid muscle on both baseline and early response scans. In all cases, the tumor region with highest k₃ also exhibited the lowest DV, typically between 0.4 and 0.6 (thus causing underestimation of total FMISO uptake), possibly indicating a mixture of severely hypoxic, cystic and/or partially necrotic tissue. In 14 lesions, TMR>1.2 was observed on early response scans whereas k₃ was <0.0026 min⁻¹ throughout the lesion.

Patient-based results are summarized in Table 4. The majority of patients (69 out of 72) were hypoxia- positive (i.e., having at least 1 lesion with TMR>1.2) on baseline scans. The normoxic three patients remained hypoxia-negative on early response scans, while hypoxia resolution was observed in 20 out of the previously hypoxic 69 patients. In 13 patients without TMR>1.2 lesions on early response scans, pharmacokinetic modeling indicated the presence of hypoxic regions. However, all 13 of these patients had at least one lesion with TMR=1.2. In 5 patients, TMR>1.2 was observed on early response scans whereas k₃ was <0.0026 min⁻¹ throughout all patients’ lesions.

Table 4.

Discordance between visual analysis and pharmacokinetic modeling for identification of patients with hypoxia (n=72 patients).

	Baseline	Early response
Patients with no hypoxic lesions according to TMR>1.2 (VA-)	3	23
Patients with no hypoxic lesions according to k3>0.0026 min−1 (PM-)	4	15
VA−, PM−	1	10
VA−, PM+	2	13
VA+, PM+	66	44
VA+, PM−	3	5

TMR - Tumor-to-Muscle Ratio, VA - visual analysis, PM - pharmacokinetic modeling

Discussion

Monitoring early treatment response by imaging studies may provide information that can be used to design response-adapted therapy strategies. However, it is imperative that the therapy-induced changes be interpreted correctly. Establishing a non-invasive and quantitative measurement of the tumor hypoxic burden is an important first step in testing the clinical relevance of hypoxia and its utility as a tumor response biomarker. While the association between pre-treatment tumor hypoxia and poor overall survival after radiotherapy has been reviewed previously [10], recent reports also suggest that FMISO imaging during the initial phase of treatment carries prognostic value for identifying HNC patients at risk from local recurrence [24, 25]. In the current study, we performed baseline and early response FMISO dynamic PET scans in 72 HNC patients to assess the added value of pharmacokinetic modeling for elucidating changes in tumor hypoxia after initiation of chemoradiotherapy.

Static metrics of FMISO images rely on the implicit assumption that FMISO distribution volume (DV) is close to unity in all tissues with dispersion not greater than what can be attributed to image noise. While we have found that DV is indeed close to 1.0 in normal non-adipose tissues in the head and neck region, including sternocleidomastoid muscle and large salivary glands, DV was observed to be non-uniform with a relatively wide spread around 1 within tumor lesions, resulting in an over- or underestimation of the tumor hypoxic fraction [18]. In adipose tissue, the DV was found to be ~0.25 (reflecting the octanol-water partition coefficient of ~0.4). The dependence of DV on tissue composition may potentially explain low intratumor DV values resulting from heterogeneous stroma that contains adipose-like tissue and partially necrotic areas with dysfunctional vasculature. Lesions with DV values greater than 1.0 were also identified (Fig. 4B), suggesting a non-specific mechanism of radiotracer accretion.

Calculation of k₃ (a time independent kinetic rate constant) from pharmacokinetic modeling of dynamic PET data results in a more direct surrogate for hypoxia-mediated entrapment than the FMISO uptake from a static scan, since the estimation of k₃ is independent of DV, tumor perfusion and imaging time post-injection. In general, changes in TBR and k₃ on a whole tumor level were well correlated. However, in 18 out of 148 lesions the total intratumor FMISO uptake decreased on the early response scan (ΔTBR ≤ −0.10) while the hypoxia parameter k₃ remained unchanged or increased (Δk₃ ≥ 0.0000 min⁻¹), in part because of a decrease in DV (ΔDV = −0.22±0.12). The correlation between k₃ and TBR was lower on early response scans compared to baseline scans, indicating the importance of also accounting for the changes in DV when interpreting FMISO static uptake. The changes in DV might be secondary to radiation-induced alteration of the soluble and insoluble components of the tumor microenvironment that affects tissue composition (e.g., leading to the development of cystic structures or necrosis; [26]). The early response scans also result in a convergence of TBR values to values around 1.0 and an increase in the number of operationally zero k₃ values, a possible and expected consequence of radiation-induced reoxygenation [27]. FMISO2 scans were performed relatively early into treatment, after patients received 1 cycle of chemotherapy and 5–10 fractions of 2Gy/fraction radiotherapy. We observed that in general, tumor shrinkage lagged behind functional change. In a recent study in a same cohort of HNC patients, Paudyal et al. reported an increase in the apparent diffusion coefficient on diffusion-weighted MRI, indicating a decrease in cellularity in the tumors weeks before any observed volume reduction [28].

The patient examples highlight the potential of FMISO pharmacokinetic modeling to elucidate microenvironmental changes that might be helpful in identifying appropriate therapeutic options. However, the added benefit of performing FMISO dPET depends on the clinical task. For example, in a clinical trial where absence of hypoxia in both primaries and lymph nodes would lead to de-escalation of therapy [6], the disagreement regarding hypoxic status of lesions between visual analysis (TMR>1.2) and pharmacokinetic modeling results (k₃≥0.0026 min⁻¹) would have affected 13 out of 72 patients, however all of these patients exhibited lesions with TMR=1.2 (i.e., if TMR≥1.2 were adopted as hypoxia threshold, there would be no disagreement between results from pharmacokinetic modeling and visual analysis for the purpose of dose de-escalation). Low DV values in these cases indicate the onset of therapy-induced necrosis, making such discordance possibly clinically inconsequential. The consequence of false positive hypoxia readings, i.e., when the nuclear medicine physician reports a TMR>1.2 but the k₃ value is near zero, is less egregious (5 out of 72 patients, Table 4). In such cases, dose de-escalation would not be recommended and conventional radiation therapy doses delivered. If hypoxia would have been targeted (e.g. via dose escalation), a consensus among the investigators would have to be reached on whether to classify patients as having hypoxic or normoxic lesions. However, the encouraging outcome results suggest that dose escalation, with the intent to improve outcome, is of lower priority in this cohort of patients. ¹⁸F-FMISO pharmacokinetic modeling might however be more useful for characterization of the response to treatments that have a direct effect on tumor vascularization and perfusion [29].

Further work is necessary to understand the clinical relevance of the disagreement between visual analysis and pharmacokinetic modeling results and to establish the reproducibility of pharmacokinetic modeling. As only 4 out of 72 patients were deceased at the time of writing (1 metastatic recurrence, 1 local and metastatic recurrence and 2 deaths from other causes; median follow-up period: 26 months), and additional 2 patients experienced local recurrence, survival analysis is presently underpowered due to a paucity of events.

Conclusion

Pharmacokinetic modeling of FMISO dynamic PET reveals a more detailed characterization of the tumor microenvironment and assessment of response to chemoradiotherapy in HNC patients than a single static image does. In a clinical trial where absence of hypoxia in primary tumor and lymph nodes would lead to de-escalation of therapy, the observed disagreement between visual analysis and pharmacokinetic modeling results would have affected patient management in <20% cases. While simple static PET imaging is easily implemented for clinical trials, the clinical applicability of pharmacokinetic modeling remains to be investigated.

Supplementary Material

259_2017_3720_MOESM2_ESM

Supplementary Figure S1. Changes in k₃ are correlated with changes in TBR.

(A) Scatterplot of changes in mean intratumor values of k₃ and TBR after 1 cycle of chemoradiotherapy, color-coded according to changes in mean intratumor DV.

(B) Scatterplot of changes in mean intratumor values of k₃ and DV after 1 cycle of chemoradiotherapy, color-coded according to changes in mean intratumor TBR.