Background
Minimal progress has been made in the management of soft tissue sarcoma (STS) over the past several years [
1]. The underlying difficulty behind this effort is multifactorial, as sarcoma is a rare disease entity representing approximately 1% of cancer incidence in the United States and possesses significant intra- and inter-tumoral heterogeneity, consisting of dozens of histologies with varying clinical behaviors [
2]. Despite this known diversity, the current standard of care is a uniform, conventionally fractionated radiation dose delivered homogenously, followed by a planned operation, regardless of histology, grade, or obvious radiographic intratumoral heterogeneity [
3‐
5]. This treatment algorithm results in local recurrence rates ranging from 10 to 20%, the majority of which occur within the radiotherapy treatment field. In addition, standard neoadjuvant radiation treatment achieves a favorable pathological response (FPR ≥ 95% cellular response) in only 8% of cases [
6]. These findings clinically support the claim that STS may be inherently radioresistant [
1,
7‐
9], which has also been described radio-biologically, with an alpha/beta ratio ranging from 2 to 6 [
10,
11].
In recent studies, we have developed the genomic adjusted radiation dose (GARD), an RT-specific metric that estimates the clinical effect of a given RT dose for an individual tumor. GARD is based on the gene expression-based radiosensitivity index (RSI), a clinically validated genomic signature of cellular radiosensitivity and the linear quadratic model [
12,
13]. In a pan-cancer analysis including 1615 patients in seven different disease sites, GARD outperformed RT dose (EQD2) by its association with clinical outcome (overall survival and recurrence risk), and its prediction and quantification of RT benefit for each individual patient [
12]. STS was found to have a broad range of radiosensitivity across various histologies, where the ideal neoadjuvant dose for highly radioresistant STS histologies is a BED
3.29 ≥97 Gy, which translates to 57.5 Gy in 25 fractions [
11]. This is consistent with the improved local control seen with neoadjuvant simultaneous integrated boost in retroperitoneal sarcoma, where 57.5 Gy in 25 fractions had a higher 5-year abdominopelvic control (96% vs. 70%,
p = 0.046), when compared to standard fractionation RT [
14].
Given its ability to quantify RT benefit, we have proposed that GARD could be utilized to design clinical trials. To support this, we have demonstrated that GARD-based model predictions for uniform dose escalation (as in RTOG 0617 for NSCLC) or uniform de-escalation in HPV + oropharynx cancer (as in HN005) align with the clinical results reported for the actual trial. There are two approaches to GARD-based clinical design: (1) Uniform GARD-optimized dose and (2) Personalized GARD-optimized dose. The first approach is focused on diseases where GARD estimates patients are either primarily under/over-dosed with RT and thus a single escalated/de-escalated RT dose is predicted to impact the overall treatment benefit for the population. In the second approach, no single RT dose is identified that can improve the overall outcome of the population and only a patient-specific personalized approach is predicted to work. GARD-optimization models propose that in STS a uniform approach would be successful.
Although, STSs are primarily considered radioresistant, gene expression and RSI variability exists both between tumors and within each tumor [
15], which may account for the differences in radiation responses to subpopulations within the tumor [
16]. In addition, tumor hypoxia [
17] might be a contributor factor to this phenotype, with radiographically identifiable regions within STS exhibiting poor perfusion and dense cellularity [
18‐
20]. From a prognostic standpoint, tumor hypoxia has been associated with a more aggressive phenotype [
21] and a higher risk of metastatic disease [
22,
23]. This association may explain the distant control and overall survival benefit observed when FPR is achieved, in addition to the expected locoregional control benefit [
24,
25], which may be in part due to the improved R0 (i.e., negative tumor at ink) resection rates in patient that achieve FPR
6 26. Therefore, efforts to personalize treatment both inter- and intratumorally are required to optimize FPR rates and, in turn, patient outcomes.
Advances in our understanding of tumor heterogeneity, radiosensitivity, and radiomics have provided the opportunity to personalize our radiation treatment planning and provide directed treatment escalation of STS, thereby improving clinical responses and outcomes. These improvements have been facilitated by advances in tumor imaging, much of which is part of the standard workup for STS, including computed tomography (CT) and magnetic resonance imaging (MRI). Less commonly used MRI sequences, such as apparent diffusion coefficient (ADC) and dynamic contrast enhanced (DCE) images, demonstrate cellularity and hypoxia within STS [
18‐
20] and are prognostic for tumor response to radiotherapy [
27‐
32] as early as week two of treatment [
28]. By using these sequences, MRI habitat analyses identify cellular subpopulations that may represent a distinct tumor biology [
3,
4,
33]. Overlaying of these sequences, which identify cellular density, hypoxia, and perfusion, has allowed for distinction between regions of viable oxygenated, viable hypoxic, and necrotic cells [
16,
29,
34,
35]. The focal dose escalations would be limited to regions within the gross tumor volume (GTV), while maintaining the same microscopic dose for the broader clinical target volume (CTV) coverage, without a significant change in dose to the surrounding normal tissue [
36]. Therefore, STS is the ideal disease to utilize genomics and radiomics to effectively personalize radiation treatment, optimizing pathologic response, R0 resection, and outcome.
In this study, we describe a prospective clinical trial to test a novel genomic-based RT dose optimization algorithm (GARD), utilizing radiomics-habitat directed targeting, to improve the clinical outcomes for soft tissue sarcoma. GARD-based clinical trial modeling proposes that selectively increasing dose (60 and 70 Gy in 25 fractions) to the radioresistant half of the tumor will triple the number of patients that experience a favorable pathological response, compared to standard of care dose (50 Gy in 25 fractions). To identify which intratumoral regions would benefit from the higher optimized doses, we integrate a radiomic habitat-based approach, directing the GARD-optimized RT dose to the cell dense and hypoxic MRI based subpopulations. This isotoxic approach is hypothesized to result in acceptable normal tissue dosing for the GARD-optimized RT dose, focally escalating radiation dose levels with the use of modern radiation techniques (e.g., simultaneous integrated boost [SIB]). The hypofractionation offers a higher biological effect on these low α/β regions, providing an avenue for safely improving FPR rates without significantly increasing toxicity, which is associated with improved local control, distant control, and overall survival [
24‐
26,
37]. With improved understandings of tumor heterogeneity, radiosensitivity, and radiomics, our goal is to personalize radiation treatment for each patient.
Aims
The primary aim of the study is to determine whether radiomic habitat directed GARD-optimized RT dose escalation can significantly increase the favorable pathologic response rate by > 3 fold over historic standard neoadjuvant radiotherapy (7.9%) to 24%. In addition to safety and feasibility of this approach, we will investigate predictors of response, radiomics-genomic correlation, circulating biomarkers of response, and the genomic heterogeneity within these habitats to create non-invasive radiomic biomarkers for further prospective validation.
Discussion
HEAT is the first clinical trial combining genomics and radiomics to personalize radiotherapy in STS, with the goal of significantly improving treatment response and outcome. This clinical trial utilizes a novel genomic-based algorithm (GARD) and radiomic-habitat directed treatment approach, to safely deliver optimized and escalated isotoxic RT doses (60 or 70 Gy in 25 fractions) to regions of high radioresistance. If successful, this trial would have implications beyond soft tissue sarcoma. We would be demonstrating for the first time that it is possible to improve the clinical outcome of RT-treated patients by integrating biological and radiologic features into clinical trial design and dose determination.
Current studies exploring neoadjuvant treatment intensification for high grade STS are investigating combinational therapies in conjunction with RT. Two such interventional studies that are currently enrolling are the phase 1 “Gemcitabine and Docetaxel With Radiation in Adults With Soft Tissue Sarcoma of the Extremities” study (NCT04037527) and phase 2 “Preoperative IMRT With Concurrent Anlotinib for Localised Extremity or Trunk Sarcoma” study (NCT05167994). An observational study out of Mayo Clinic, “An Imaging Agent (Fluorodopa F 18) With Positron Emission Tomography/Magnetic Resonance Imaging for Assessing Treatment Response in Patients With High-Grade Soft Tissue Sarcomas” (NCT05560009), is evaluating novel imaging techniques to evaluate RT treatment response. The HEAT trial differs from other contemporary trials because it incorporates genomic and radiomic understanding of STS to craft an individualized and GARD-based treatment to optimize RT dose to radioresistant intratumoral habitats while maintaining clinically acceptable doses to nearby normal structures for high-grade STS.
Understanding STS intra-tumoral heterogeneity and radiosensitivity opens the potential for a more personalized neoadjuvant RT paradigm that utilizes SIBs for GARD-optimized dose adjustments to radioresistant areas of the tumor
18–20,27−32. Over the past decade, the convergence of radiomics and genomics has given rise to radiogenomics, a discipline that explores the relationship between radiologic phenotypes and genomic intratumor heterogeneity [
43‐
45], which posits that distinct radiological features of tumors are influenced by their underlying biology at the cellular and molecular levels. MRI has demonstrated an ability to identify tumor biology, with radiomic features often predictive of specific genomic markers. Intratumoral heterogeneity can be identified with radiomic habitats [
16,
29,
34,
35] that are derived by overlapping different imaging sequences to pinpoint unique radiographic regions in tumors [
27‐
32]. These habitats are then able to serve as biomarkers to predict prognosis and treatment response [
24,
25]. The HEAT trial takes advantage of these habitats and uses them to guide GARD-based dose escalation and optimization. Additionally, this approach helps to limit toxicity associated with dose escalation, by only targeting areas known to be the most radioresistant within the actual gross tumor, while allowing the edges of the field that approach or abut critical OARs to receive typical doses as with standard neoadjuvant RT in STS. We hypothesize that this radiomically directed GARD-based dose escalation and optimization will increase the FPR rate by threefold, from a historic 7.9% to an estimated 24.3%.
Prior clinical trials employing uniform dose adjustments (e.g., RTOG 0617, HN005) have not demonstrated improved outcomes for RT-treated patients. In RTOG 0617, uniform RT dose escalation to 74 Gy resulted in decreased OS for NSCLC patients, whereas de-escalation to 60 Gy for HN005 also negatively impacted HPV + oropharyngeal patients [
46]. Although clinically unexpected, post-hoc GARD modeling predicted that uniform dose changes applied to all patients would be unsuccessful, whereas pathogenomic-based patient selection for dose escalation/de-escalation would improve patient outcomes. HEAT is the first clinical trial that uses a biological basis for RT dose adjustments, incorporating these predictions into the statistical design of the trial, thus marking a new era for radiation oncology. STS, as identified by GARD, is a disease where these dose optimizations would result in improved treatment response and disease control. The combination of GARD with an MRI radiomic-based habitat identification strategy allows for an intelligent isotoxic SIB approach to these radioresistant intratumoral regions without significant toxicity risk. This is unlike other disease sites (e.g., lung, breast, oropharynx cancer) where only a personalized approach prior to dose adjustments is predicted to impact outcome. GARD predicts that a large treatment benefit would be realized if the dose is selectively increased to 60 or 70 Gy in 25 fractions, a hypothesis being prospectively tested in this trial.
Image sequences commonly obtained for STS workup include CT scans and MRI. Additional functional MRI that may provide insight into tumor biology, include diffusion weighted imaging (DWI), apparent diffusion coefficient (ADC), and dynamic contrast-enhancement (DCE) sequences. DWI is able to evaluate the extent and direction of random water motion in tissue [
47], which can aid in determining cellularity, aggressiveness, and response to therapy [
30]. An apparent diffusion coefficient (ADC) map can be generated from the DWI by ignoring the contribution of the microcirculation signal. The ADC, or the measured diffusion in tissue, is dependent on the tortuosity of the extracellular space, with lower ADC values associated with increased tumor cellularity and architectural distortion, commonly seen with high tumor grade [
48,
49]. A DCE sequence measures rate and magnitude of perfusion and can aid with differentiation between regions of hypoxia, necrosis, and normal perfusion, which correlates to radiographic cellularity and hypoxia in STS [
18‐
20,
50]. By overlaying DCE and ADC sequences, we can identify radiomic habitats based on cellularity and hypoxia, which offer a radiographic representation of the innate tumor biology, allowing us to personalize radiation dose levels within the tumor based on the subclonal region’s phenotype. In addition, tumor hypoxia is more often associated with a more aggressive phenotype [
21], a higher risk of metastatic disease [
2223], and resistance to radiation [
17].
Furthermore, we also aim to improve understanding of STS radiobiology. For this effort, we will be performing stereotactic biopsies prior to treatment within each habitat and placing a marker. This approach was developed to allow for precise genomic and tissue level evaluation of these different habitat regions and how they respond to different doses. It will allow a comparison between the habitats in terms of their genomics, pathways activated, GARD score, and contribution to circulating tumor DNA (ctDNA). Then each habitat will be assessed to determine individual response to these regions that can be associated with the dose delivered to the given region. Additionally, mpMRIs will also be performed mid treatment during weeks two to three to biologically adapt treatment based upon initial treatment response, and again repeated after the completion of radiotherapy. When treated on the MRL, multiparametric imaging will also be conducted on the MRL to capture both first and second order radiomic features taken before, during and after radiation, as well as daily TRUFI sequences. Along with each mpMRI, pre-treatment, mid-treatment, and post-treatment peripheral liquid biopsy samples will be collected to investigate immunological markers (e.g., peripheral blood mononuclear cells and myeloid derived suppressor cells) and determine if any ctDNA biomarkers are identifiable and how they correlate with treatment response.
Radiomic analysis of the mpMRI data from both the diagnostic and MRL images will be evaluated for their potential to predict tumor response to radiation. Due to the differences in field strength and resolution between diagnostic MRI and MRL images, we will calibrate these images and explore conversion factors, which could enhance the clinical utility and prognostic potential of MRL images. Radiomic data will be analyzed in conjunction with blood, plasma, and tumor genomic data to identify potential radiomic and genomic biomarkers. These biomarkers hold considerable potential to guide future studies to improve personalized radiation therapy for STS, such as improved patient selection, more sophisticated biologically guided RT techniques, and predict patients at high-risk for treatment failure who may benefit from further adjuvant therapies.
Overall, the HEAT trial takes full advantage of current understanding within various fields (i.e., radiomics, genomics, and MRI-guided RT) to push beyond traditional one-size-fits-all RT paradigms. It represents an important step towards biologically and radiographically guided adaptive radiation therapy. If successful, the HEAT trial would provide the first evidence that GARD-based dose adjustments can improve the clinical outcome of patients as predicted, improve our understanding between the relationship between radiomics and genomics, and would open a new era in radiation oncology leading to genomic and radiomic-based clinical trials and practice. In addition, the insights gathered may also enable the identification of novel radiomic and genomic biomarkers that may lead to a refinement of current personalized strategies for STS patients.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.