Histology Specific Molecular Biomarkers: Ushering in a New Era of Precision Radiation Oncology

Abstract

Histopathology and clinical staging have historically formed the backbone for allocation of treatment decisions in oncology. Although this has provided an extremely practical and fruitful approach for decades, it has long been evident that these data alone do not adequately capture the heterogeneity and breadth of disease trajectories experienced by patients. As efficient and affordable DNA and RNA sequencing have become available, the ability to provide precision therapy has become within grasp. This has been realized with systemic oncologic therapy, as targeted therapies have demonstrated immense promise for subsets of patients with oncogene-driver mutations. Further, several studies have evaluated predictive biomarkers for response to systemic therapy within a variety of malignancies. Within radiation oncology, the use of genomics/transcriptomics to guide the use, dose, and fractionation of radiation therapy is rapidly evolving but still in its infancy. The genomic adjusted radiation dose/radiation sensitivity index is one such early and exciting effort to provide genomically guided radiation dosing with a pan-cancer approach. In addition to this broad method, a histology specific approach to precision radiation therapy is also underway. Herein we review select literature surrounding the use of histology specific, molecular biomarkers to allow for precision radiotherapy with the greatest emphasis on commercially available and prospectively validated biomarkers.

Introduction

Oncologists have historically been limited to utilizing clinical features such as histopathology and staging information to guide treatment decisions. Although this has been a practical and fruitful approach for decades, it is clear that even among patients with similar clinical cancer features, heterogeneity in response to treatment and outcomes exist. The paradigm of more precision therapy in cancer began to evolve with the identification and targeting of driver mutations and a more personalized approach to treatment became possible^1–5. As efficient and affordable next-generation sequencing (NGS) has become available, more personalized approaches to oncologic therapy have garnered considerable excitement and become increasingly utilized⁶.

Most of the advancements within precision oncology currently in clinical practice have been in relation to systemic therapy. Within the field of radiation oncology, advancements in precision of treatments have predominately been with respect to improvements in localization^7–10 and conformality^11–14. Dosing and fractionation however, is still largely determined by histology and extent of disease (macroscopic vs microscopic) while maintaining safety to relevant organs at risk. Several clinical trials in various disease sites have evaluated dose escalation/de-escalation strategies in an effort to identify the optimal therapy to balance tumor control with toxicity. Although this has allowed for a general identification of safe and therapeutic doses, it is evident from clinical outcomes there is significant heterogeneity within patients with matched tumor type and stage.

Gaining a better understanding of the prognostic and predictive value molecular alterations may allow for greater personalization of radiation tailored to the radiosensitivity of a specific tumor. The genomic adjusted radiation dose/radiosensitivity index^15,16 is one such emerging focus of study that holds promise to improve precision radiation dosing with a pan-cancer approach and is the subject of a separate review within this special issue. Herein, we review current select literature of histology specific molecular biomarkers that may guide precision radiotherapy (RT) with a greater focus on breast and prostate cancer due to the presence of commercially available and prospectively validated biomarkers.

Breast Cancer

Several predictive and prognostic molecular classifiers have been developed within both non-invasive (ductal carcinoma in situ, DCIS) and invasive breast cancer^17–21. For DCIS, the DCISionRT is a commercially available and clinically validated biologic risk signature designed to predict benefit to adjuvant RT following Breast Conserving Surgery (BCS) in DCIS^22,23. The score combines seven protein tumor markers in combination with four clinicopathological factors. Wärnberg et al²⁴. evaluated 504 patients enrolled on the SweDCIS²⁵ randomized trial (BCS vs BCS+RT) who underwent DCISionRT testing. Patients with an elevated decision score (>3) experienced a significant reduction (9.3% absolute decrease) in ipsilateral invasive recurrence with adjuvant RT whereas patients with a low-risk score experienced no significant benefit. Further analysis identified while a cutoff of 3 did not meet statistical significance to predict for RT benefit (p interaction=0.09), a cutoff value of 2.8 was predictive (p interaction = 0.04). This dataset was further validated in a larger dataset of 926 patients where three biosignature risk groups were identified: low risk (score of < 2.8); elevated risk (score > 2.8 without residual risk subtype) and residual risk (score > 2.8 with residual risk subtype)²⁶. There was no significant difference in ipsilateral breast tumor recurrence rates in the low-risk group with or without RT; in the elevated risk group, rates were decreased with the addition of RT (p<0.001); and in the residual risk group, rates were also decreased with the addition of RT (p<0.001) and high overall for recurrence even with the addition of RT with 10-year rates of 14.7%. These results, although promising, should be interpreted cautiously until this biomarker can be more definitely validated with at least a prospective randomized cohort or ultimately in an integral biomarker designed trial²⁷.

The Oncotype Dx and Mammaprint are two commercially available and prospectively validated genomic classifiers (GCs) for patients with invasive breast cancer. The Oncotype Dx incorporates expression of 21 (16 cancer related and 5 reference) genes through reverse-transcriptase polymerase chain reaction (RT-PCR). The score ranges from 0–100 and stratifies patients into “low”-, “intermediate”-, and “high”-risk and is prognostic for 10-year distant recurrence²⁰. Since its development, the Oncotype Dx recurrence score (RS) has been incorporated into randomized clinical trials to guide systemic therapy decision making. The TAILORRx^28,29 and RxPONDER³⁰ trials have demonstrated the clinical utility of the Oncotype Dx RS for determining benefit of adjuvant chemotherapy in patients with lymph node negative and positive disease, respectively. Similarly, Mammaprint is a GC that incorporates expression of 70 genes through microarray analysis^31,32. The score ranges from −1.0 to 1.0 and categorizes patients into “low’- and “high”-risk of recurrence. This GC was incorporated into the MINDACT trial³³ which demonstrated its ability to guide adjuvant chemotherapy within clinically high-risk, genomically low-risk patients. In addition to GCs, molecular subtypes were pioneered within breast cancer. The PAM50 algorithm is a clustering-based GC based on microarray expression of a set of 50 genes. Tumors are categorized into “intrinsic” cellular subtypes including luminal A and B, along with HER2 enriched and basal molecular subtype which have demonstrated improved prognostication over clinical factors and immunohistochemistry^18,19,34,35.

PAM50 molecular subtypes have been evaluated for guidance of adjuvant radiation following BCS. Nguyen et al., who used receptor status to define subtypes, demonstrated that among patients treated with BCS and adjuvant radiation, those with HER2 and basal subtypes had significantly higher rates of local recurrence (luminal A: 0.8%, luminal B: 1.8%, HER2: 7.1%, basal: 8.4%)³⁶. More recently, results from an expanded cohort demonstrated similar findings³⁷. Although several additional studies have generally demonstrated higher rates of LRR with triple-negative disease³⁸, others have demonstrated triple-negative disease is associated with higher risk distant recurrence and death but not LRR^39,40. Given the very low rates of recurrence amongst Luminal A tumors, the LUMINA trial evaluated the omission of RT following BCS in women >55 years old with early-stage Luminal A tumors, defined as estrogen receptor (ER) expression ≥1%, progesterone receptor (PR)>20%, HER2-negative and Ki67≤13.25%. The 5-yr results of this trial were recently presented, demonstrated a low risk of recurrence with 5-yr local recurrence of 2.3% which is approximately the risk of a new contralateral breast cancer⁴¹.

Due to very low rates of local recurrence within early stage, ER+ breast cancer treated with BCS followed by adjuvant RT and estrogen therapy, there has been significant interest in treatment deintensification. Two large randomized clinical trials CALGB 9343⁴² and PRIME II⁴³ demonstrated that among elderly patients with early-stage ER+ breast cancer receiving adjuvant estrogen therapy, radiation can be safely omitted without any detriment to 5-year overall survival. Despite these promising results, both of these trials demonstrated estrogen therapy alone was associated with inferior locoregional control. Notably, neither of these trials stratified patients based on genomic risk score which may better aid in selecting patients in whom omission of adjuvant radiation is most appropriate. Chevli et al. performed a retrospective study of the National Cancer Database (NCDB) of elderly women (>70 years old) with pT1N0, ER+/PR+, HER2− breast cancer treated with BCS and estrogen therapy with or without adjuvant radiation and stratified patients based on their Oncotype Dx RS⁴⁴. Among 11,891 patients included in the study, adjuvant radiation demonstrated an improvement in 5-yr OS in patients with intermediate or high RS (91% vs 87%, p=0.003) but not low RS (91% vs 89%, p=0.61). This will be further evaluated by NRG BR007 which will randomize 1670 patients with stage I, hormone sensitive, HER2− breast cancer with Oncotype RS ≤ 18 to BCS + estrogen therapy with or without RT (whole breast + boost, partial breast irradiation, or accelerated partial breast irradiation).

Although omission of radiation is being extensively evaluated for low-risk patients, GCs may also identify a subset of patients in whom treatment intensification is necessary. Kim et al. retrospectively evaluated 339 lymph node negative, ER+/HER2− patients enrolled on Korean Radiation Oncology Group 19–06 treated with BCS followed by whole-breast irradiation and estrogen therapy who underwent Oncotype Dx testing⁴⁵. Within this cohort, all patients who developed a locoregional recurrence (LRR) had a high RS resulting in a 5-year LRR of 7.3% in this subgroup. Given this relatively high rate of LRR the authors conclude the potential need for regional nodal irradiation (RNI) within this higher risk subset. Mamounas et al. examined the association of LR with Oncotype Dx RS in ER-positive, node-negative breast cancer patients on the National Surgical Adjuvant Breast and Bowel Project (NSABP) B14 and B20 clinical trials⁴⁶. No patients received targeted regional nodal irradiation on these trials. These results demonstrated that the Oncotype Dx RS was significantly associated with locoregional recurrence. The incorporation of RNI to this subset however should be prospectively evaluated before becoming routine clinical practice.

Within node positive disease, Laurberg et al. evaluated the relative benefit of adjuvant RT following mastectomy for different molecular subtypes of patients enrolled on the British Columbia⁴⁷ and Danish Breast Cancer Cooperative Group 82b⁴⁸ trials. This identified patients with Luminal A tumors achieved the greatest improvement in locoregional control when treated with adjuvant RT (HR 0.12, 95%CI 0.02–0.60)⁴⁹. In a similar analysis to those in node-negative, Mamounas et al. demonstrated patients with node positive, ER-positive breast cancer in the NSABP B-28 trial, the Oncotype Dx RS was a significant predictor of locoregional recurrence in both univariate and multivariate analysis⁵⁰. Similarly, a retrospective analysis of the Southwest Oncology Group (SWOG) S8814 study of patients with ER+, node-positive breast cancer found that a higher Oncotype Dx RS was prognostic for locoregional recurrence⁵¹.

Currently, there have been several attempts to develop a radiation-specific molecular classifier to identify patients with breast cancer who may benefit most from RT⁵². Sjöström et al. identified a novel 27-gene expression Adjuvant Radiotherapy Intensification Classifier (ARTIC) predictive of response to radiation in three publicly available cohorts⁵³. This classifier was subsequently validated following transcriptomic profiling of patients enrolled on the SweBCG91-RT trial (randomized patients with BCS +− RT). Patients with low ARTIC score demonstrated a significant 10-yr LRR benefit from adjuvant RT (21% vs 6%: HR=0.33, p<0.001) whereas those with high ARTIC (32% vs 25%, HR=0.73, p=0.23) did not. This represents the first, breast specific, predictive genomic signature for adjuvant radiation validated within a phase III trial. Work by Speers et al. developed a radiation sensitivity signature (RSS) based on 51 genes using clonogenic survival assays and a training dataset 343 breast cancer patients with transcriptomic evaluation ⁵⁴. The RSS was further validated in an independent clinical dataset and demonstrated it to be prognostic for both LRR (HR =5.3, p<0.001) and OS (HR=1.8, p<0.05), outperforming intrinsic breast cancer subtype. Cui et al. also developed and validated an intrinsic radiosensitivity signature to predict local recurrence after RT⁵⁵. Within a validation cohort of 1,439 patients, patients classified as radiosensitive demonstrated an improvement in disease-specific survival (DSS) with the addition of RT (HR=0.68, p=0.059). Conversely, within the radioresistant cohort, radiation was associated with inferior DSS (HR=1.53, p=0.059). This radiosensitive signature was found to be predictive of benefit to radiotherapy (P_interaction=0.007).

Finally, molecular classifiers have been evaluated for determination of response to radiation fractionation. Among patients with lymph node negative breast cancer receiving adjuvant whole breast radiation following BCS Bane et al. evaluated whether molecular subtypes were predictive of response to hypofractionated RT⁵⁶. This work evaluated 989 patients who were enrolled on the Hypofractionation Whole Breast Irradiation Trial⁵⁷ (randomized patients to either hypofractionated WBI 42.5Gy/16 fractions vs conventional WBI in 50Gy/25 fractions) and underwent molecular subtyping. No significant interaction was identified between fractionation and subtype on local recurrence risk. This however was limited by few events in the cohort. To address these limitations Lalani et al. similarly evaluated the effect of molecular subtype on response to radiation fractionation⁵⁸. This work identified 5868 patients with stage I-III breast cancer who underwent molecular subtyping and received either hypofractionated or conventional fractionation RT. This work also demonstrated no differences in 10-year local recurrence-free survival between hypo- or conventional fractionation within any subtype. These findings are consistent with the American Society of Radiation Oncology’s guidelines on whole breast radiation indicating dose-fractionation should not currently be dependent upon tumor biology⁵⁹.

Prostate Cancer

Localized prostate cancer has several commercially available genomic biomarkers that aim to improve risk stratification over clinical factors alone. One such commonly utilized and well validated classifier, the Decipher GC is a commercially available 22-gene classifier that utilizes a whole-transcriptome oligonucleotide analytically validated microarray platform. The GC is comprised of RNA expression of genes related to androgen receptor signaling, cell proliferation, differentiation, motility and immune modulation⁶⁰. The GC reports a score ranging from 0–1 estimating the risk of adverse pathologic features (Grade group 3–5, pT3b-T4, lymph node involvement) at radical prostatectomy (RP), distant metastasis, and prostate cancer specific mortality.

Active surveillance (AS) is currently a favored option for the management of patients with very low- or low-risk prostate cancer, as it avoids/delays side effects associated radical treatment without affecting prostate-cancer specific survival⁶¹. Although excellent salvage treatment options exist in the setting of disease progression (including either RP or RT), there remains a potentially increased risk of development of distant metastasis compared to those undergoing upfront radical treatment⁶¹. Among men with very low-, low-, or favorable intermediate-risk prostate cancer, Kim et al. demonstrated the Decipher biopsy GC can predict for adverse pathologic features per 10% increase in score and demonstrated a negative predictive value of 96% when Decipher score was ≤ 0.2⁶². These results indicate the GC may be used to best select patients for which AS is most appropriate and who may benefit from early radical intervention. In patients with favorable intermediate-risk disease, in which AS is controversial, only those with Decipher high-risk tumors (score >0.6) had increased risk of adverse pathology upon radical prostatectomy. This may provide greater reassurance for low GC favorable intermediate-risk patients on surveillance⁶³. Among patients managed with active surveillance, GC has been shown to be associated with short-term biopsy Gleason up-grading⁶⁴. Analysis of the prospective state-wide MUSIC registry has shown with real-world data, a lower GC was associated with increased utilization of AS⁶⁵ and after adjusting for clinical risk factors, a high-risk GC score was independently associated with shorter time to radical therapy in these men⁶⁶. Unfortunately, there is currently no consensus on what threshold score should be utilized to guide management, emphasizing the importance of shared decision making between the patient and physician.

Outside of identifying patients in whom omission/delayed RT is safe, the GC has been proposed to identify patients at higher risk of developing metastatic disease following radical treatment. Jairath et al. reported a systematic review of prospective and retrospective studies evaluating the Decipher GC on biopsy tissue.⁶⁷ These studies consistently demonstrated increased prognostication of developing metastatic disease across low-, intermediate- and high-risk prostate cancer treated with either radical prostatectomy or RT +/− androgen deprivation therapy (ADT) while clinical and pathologic features were inferior in this regard ^68–73. Additionally, the incorporation of the GC to the regression models based on either NCCN risk grouping, CAPRA⁷⁴, or Stephenson⁷⁵ models demonstrated significantly improved AUC and C-index metrics^68–72.

Several ongoing prospective clinical trials are aiming to incorporate the GC for greater precision in the management of localized prostate cancer. Two NRG studies GU009 and GU010 are aim to de-intensify or intensify high- and unfavorable intermediate-risk prostate cancer, respectively, based on GC risk. GU009 will randomize 2,478 patients with high-risk prostate cancer and those with: (i) low/intermediate GC risk (score ≤ 0.85) to either standard of care with RT + 24 months ADT versus RT + 12 months ADT (Deintensification arm); or, (ii) high GC risk (score >0.85) to either same standard of care versus the addition of apalutamide (Intensification arm). GU010 will randomize 2,050 patients with unfavorable intermediate-risk prostate cancer and those with: (i) low GC risk (score <0.4) to either standard of care with RT + 6 months ADT versus RT alone (Deintensification arm); or, (ii) higher GC risk (score ≥ 0.4) to same standard of care versus the addition of darolutamide. The genomics in Michigan to Adjust Outcomes in Prostate cancer (G-Major) trial is a randomized trial enrolling 900 patients with newly diagnosed favorable intermediate-risk prostate cancer to either standard of care versus integration of GC to further guide management. The investigators hypothesize a greater proportion of patients will be managed with active surveillance within the GC arm. In the postoperative setting, EA8183 aims to enroll 810 high-risk men with adverse pathology after RP and a GC>0.6 is used further as a selection factor to randomize patients to standard of care RT+ADT versus the addition of darolutamide. Notably, these trials are focused on guiding precision systemic therapy rather than radiotherapy dosing.

Pioneered within breast cancer, molecular subtyping has also been applied to prostate cancer as a similarly hormone driven malignancy. As noted above, the PAM50 algorithm is a clustering-based genomic classifier based on the expression of a set of 50 genes categorizing patients into “luminal” and “basal” molecular subtypes. Applying this classifier to 3782 localized prostate cancer samples, Zhao et al. segregated tumors into three distinct subtypes including luminal A, luminal B, and basal type⁷⁶. As in breast cancer, luminal subtypes have increased hormone receptor expression and downstream signaling. Given the very recent application of molecular subtyping of prostate cancer, there is limited data on how these subtypes may influence response to RT. It is presently unknown whether molecular subtypes experience differential response to total dose of radiation or to the radiosensitizing effects of ADT.

In the post-prostatectomy setting, randomized data has demonstrated radiation improves biochemical progression-free survival among patients with positive margins, extracapsular extension or seminal vesicle invasion^77–79. The Post-operative Radiation Therapy Outcomes Score (PORTOS) is a 24 gene signature that estimates distant metastases risk after adjuvant RT. This score incorporates expression of genes implicated in DNA damage repair and response to radiation. In a validation cohort of 330 patients, those treated with radiotherapy had a decreased incidence of distant metastases within the high PORTOS group (4% vs 35%: HR 0.15, p=0.002) but not the low PORTOS group (32% vs 32%: HR 0.92, p=0.76) with a significant interaction⁸⁰. Given these results, PORTOS is the only clinically validated biomarker predictive for response to RT within prostate cancer. Prospective studies have aimed to evaluate integration of genomic biomarkers into management decisions. Evaluating the Decipher GC in the post-prostatectomy setting, Marascio et al. evaluated two prospective registries (clinical utility cohort and clinical benefit cohort) of patients with prostate cancer with adverse pathology treated between 2014 and 2019. GC testing altered treatment recommendations in 39% of patients in the clinical utility cohort. Furthermore, patients with high GC risk experienced significantly improved 2-year PSA recurrence with adjuvant RT (3% vs 25%) in the clinical benefit cohort. This however was contrasted by no improvement in 2-year PSA recurrence (0% vs 2.8%) with low or intermediate risk score⁸¹. Further prospective studies similarly demonstrated the utility of the GC in guiding treatment decisions with a number needed to treat ranging from 1.5–4 patients to change management, most commonly in patients with high GC risk^82,83.

Aiming to avoid/delay the morbidity associated with adjuvant RT, several trials have compared adjuvant versus early salvage RT (defined as PSA 0.1–0.2)^84–86. The ARTISTIC meta-analysis of these trials demonstrated no improvement in event free survival with adjuvant RT compared with early salvage⁸⁷. Given these findings, early salvage has become a common management strategy among these patients. Feng et al. demonstrated the Decipher GC was independently associated with distant metastases (HR 1.17, 95%CI 1.05–1.32), prostate cancer specific mortality (HR 1.39, 95%CI 1.20–1.63), and OS (HR 1.17, 95%CI 1.06–1.29) in patients treated on RTOG 9601 (salvage RT +/− antiandrogen therapy in recurrent prostate cancer)⁸⁸. These results demonstrate that the Decipher GC has the potential for identifying patients who are at highest risk for disease progression and therefore may have the greatest benefit to aggressive early management. Feng et al. also demonstrated a nearly 3-fold greater benefit to OS with the addition of ADT in intermediate/high- (8.9% 12-OS benefit) versus low-risk (2.4% 12-OS benefit) GC. This is particularly useful as there remains disagreement in the field as to who should be offered concurrent ADT along with salvage RT, particularly for early salvage RT ^89–91. Decipher GC has also been validated within the SAKK 09/10 trial⁹² which evaluated dose-escalated salvage RT without concurrent ADT. Dal Pra et al evaluated 226 patients enrolled on the SAKK 09/10 trial who underwent GC testing⁹³. This work identified patients with high GC experienced worse biochemical and clinical progression compared to those with low-intermediate GC. Notably however dose escalation did not appear to benefit neither the overall, low/intermediate, nor high GC cohorts.

Similar to the upfront setting, several randomized clinical trials are currently enrolling to gain a better understanding of how these GCs can be incorporated in the clinic. NRG Oncology GU006 (BALANCE) is a 324-patient phase II randomized biomarker study stratified by the PAM50 classifier of salvage RT +/− the next-generation anti-androgen apalutamide. The genomics in Michigan impacting observation or Radiation (G-MINOR) trial has enrolled 356 patients treated with radical prostatectomy with adverse pathologic features and undetectable post-op PSA randomized to either receive Decipher GC versus standard of care to evaluate the proportion of patients that receive adjuvant therapy within each group and ultimately differences in both patient-reported- and longer-term oncologic outcomes.

Central Nervous System

High grade glioma continues to represent a therapeutic challenge with patients experiencing poor prognosis despite tri-modal therapy with surgery, radiation and systemic therapy. The WHO recently updated CNS tumor classification in 2021 which continued to expand the incorporation of molecular classification within glioma⁹⁴. The full expanse of these molecular prognostic features is well beyond the scope of this review; however, emerging data has highlighted the marked molecular heterogeneity of high-grade gliomas can impact treatment outcomes⁹⁵. MGMT promoter methylation status is a validated predictive biomarker for response to temozolomide for patients with glioblastoma and has become a mainstay in clinical practice. Likewise, presence of 1p19q co-deletion is highly of response to chemotherapy for patients with a Grade 3 oligodendroglioma, and IDH mutational status predictive of response to temozolomide in patients with a Grade 3 glioma^96,97. In line with the incredible clinical utility of such biomarkers for systemic therapy, there has been heavy interest in developing a molecular signature to predict sensitivity to radiotherapy. Feng et al.⁹⁸ developed and validated a radiosensitivity signature based on 5 methylation probes. This signature was built on 50 patients with primary glioblastoma who underwent RT after surgery. Patients were classified into a radiosensitive and radioresistant cohort and differentially expressed methylation probes were evaluated to create a prediction signature. The signature included methylation status of the promoter region of CCDC65, RCSD1, GNMT, ENPP2, and GLIPR1L1 with functions related to DNA mutation and repair. This signature was subsequently validated on the TCGA and CGGA database demonstrating significantly worse OS and PFS with a radioresistant signature. This signature may provide future insight for dose escalation within this radioresistant population.

Although outcomes with low grade glioma (LGG) are relatively good, there remains controversy regarding the need for adjuvant RT. Li et al trained and validated two gene signatures (radioresistance and chemoresistance score) on 942 patients with LGG from the TCGA and CGGA databases⁹⁹. These scores were then combined to create 4 subclasses (radiosensitive/chemosensitive, radiosensitive/chemoresistant, radioresistant/chemosensitive, radioresistant/chemoresistant). This provides an early framework for therapeutic precision however further clinical work is needed to validate these subclasses as a paradigm to guide adjuvant therapy for this patient population.

Within pediatric medulloblastoma, transcriptional profiling studies have demonstrated 4 distinct molecular subgroups including Wnt, Shh, Group 3, and Group 4 with distinct clinical behavior¹⁰⁰. ACNS0331 was a Children’s Oncology Group Phase III radiation de-intensification trial randomizing patients to smaller radiation boost volume (posterior fossa vs involved field) and low-dose craniospinal irradiation (CSI) (23.4Gy vs 18Gy) in children with average risk medulloblastoma¹⁰¹. Within the entire cohort, low-dose CSI demonstrated inferior event-free (71.4% vs 82.9%) and overall-survival (77.5% vs 85.6%). Notably however this difference was predominately driven by patients with molecular Group 4 disease. Given the excellent outcomes of patients within Wnt subgroup, multiple clinical trials evaluating low-dose CSI within this cohort are underway (NCT01878617, NCT02724579, NCT02066220).

Head and Neck Cancer

Radioresistance has long been a barrier to the successful treatment of head and neck cancer (HNSCC) and improving the therapeutic ratio via either altered fractionation or radiosensitizing chemotherapy plays an important role in improving local control^102,103. However, the explosive growth of human papillomavirus (HPV) positive oropharyngeal cancer over the past decade and the improved survival in this disease belies its distinct biology compared to its HPV negative counterpart. Thus, distinct and biology-driven approaches to each are required.

A detailed discussion of the biology of HPV-driven HNSCC is beyond the scope of this review, however what has been demonstrated is generally HPV positive (HPV+) HNSCC, or more specifically oropharyngeal cancer, is highly sensitive to DNA-damage based therapy such as radiation and cytotoxic chemotherapy. There are a multitude of hypotheses hat have been proposed to explain this effect, which are probably all operative to some degree. For example, it is known that the HPV E6 oncoprotein represses p53 – which is almost invariably wild type in HPV+ HNSCC relative to the high proportion of TP53 mutations seen in HPV negative HNSCC – which is involved in the complex process of HPV-related transformation¹⁰⁴. Some studies indicate that radiation can partially remove this inhibition of p53, in turn leading to p53-driven cell death^105–107. Separately, several studies have linked the presence of p16, which is the clinical surrogate for HPV positivity in the oropharynx¹⁰⁸, to a repression of homologous recombination via several mechanisms, potentially involving a non-canonical signaling cascade to ubiquitin-specific protease 7 (USP7) and TRIP12, the former being targetable^109–111. Regardless of mechanism, the high degree of responsiveness of HPV+ HNSCC, coupled with the significant toxicity of concurrent chemoradiation in the head and neck led to multiple trials in recent years to reduce the intensity of therapy with somewhat mixed results.

Broadly, de-escalation strategies can be divided into those that are focused on reducing radiation dose, replacing cisplatin with other agents or adding surgical resection to the treatment package to allow for risk-adapted therapy. The former strategy was perhaps best exemplified by NRG HN002, which was a randomized phase II trial of patients with HPV+ OPSCC which randomized 360 patients to either 60 Gy over 6 weeks with concurrent cisplatin or 60Gy over 5 weeks without cisplatin for definitive treatment. Both arms were compared against a superiority endpoint of ≥85% 2-year PFS, with the “winner” proceeding for further evaluation¹¹². While outcomes for both arms were excellent, only the concurrent arm met this pre-determined endpoint and ultimately proceeded to evaluation in NRG HN005, which compares 70Gy in 6 weeks (6 fractions/week) with concurrent cisplatin vs 60 Gy in 6 weeks (5 fractions/week) with concurrent cisplatin vs 60Gy in 5 weeks (6 fractions/week) combined with nivolumab.

While the potential of replacing cisplatin, which is a highly toxic radiosensitizer, with an alternate potentially less toxic agent is exciting, the results from large clinical trials have largely been disappointing. Two large, phase III trials RTOG 1016 and DeESCALATE both sought to replace cisplatin with cetuximab in combination with 70 Gy in 35^113,114. Unfortunately, in both studies, the cetuximab arm had a numerically inferior overall survival with no clinically useful reduction in toxicity. Moreover, the combination of radiation and concurrent immunotherapy has fared little better, with two recent trials exhibiting no benefit to concurrent treatment^115,116. While additionally biologic agents are being evaluated to potentially replace cisplatin, none are in large scale clinical trials.

A third option to potentially de-escalate therapy for HPV+ HNSCC is to utilize surgery, specifically transoral robotic surgery (TORS), in an attempt to select for patients that can be treated with uni- or bi-modality therapy. The largest trial to date to examine this concept is ECOG 3311, a randomized phase II multi-arm trial which evaluated standard (60Gy) vs low dose RT (50Gy) in resected early-stage HPV+ OPSCC with intermediate risk features. Although not powered to detect a difference between these arms, no significant difference in 2-yr PFS was detected between standard (96%; 90%CI 92.8–99.3) and dose reduced (94.9%; 90%CI 86.2–95.4) RT¹¹⁷. A sperate approach was utilized in MC1273, a single-arm phase II trial in which patients with 30 Gy in 20 fractions delivered BID with weekly docetaxel (intermediate risk) along with a simultaneous integrated boost to 36 Gy in 20 fraction BID to nodal levels with extranodal extension (high risk). This trial demonstrated a similar 2-yr PFS of 91.1% with no grade 3+ toxicity at 1 year post RT¹¹⁸. However, these data must be examined in the context of the ORATOR, which compared RT+/− cisplatin versus a TORS based approach¹¹⁹. While the investigators found no differences in outcome, there was a statistically – but not clinically – significant benefit in patient reported outcome. Further stratification within this population has demonstrated defects in tumor necrosis factor receptor-associated factor 4 (TRAF3) and cylindromatosis lysine 63 deubiquitinase (CYLD)¹²⁰ are associated with improved survival in and may hold promise for more targeted de-escalation strategies.

Regardless of de-escalation strategy, the road to de-escalation in HPV+ HNSCC, or escalation of treatment in HPV negative HNSCC for that matter, is hamstrung by a lack of clinically relevant biomarkers. While in in HPV negative, HNSCC a wide range of tumor specific markers ranging from alterations in individual genes such as TP53, CASP8, CREBBP, FAK, and EP300^121−125, to expression of specific miRNA profiles¹²⁶, to potential repression of the immune function in the tumor microenvironment^127,128 have all been linked to clinical radioresistance, these biomarkers have not been validated in a prospective fashion. Thus, they remain interesting but not clinically actionable.

Biomarkers for HPV+ disease monitoring, at least in the definitive setting, is somewhat further developed some of which capitalize on the use of cell free DNA (cfDNA) technology. While several tumor-derived signatures are available, largely related to predicting the function of HPV or p16 in some manner¹²⁹, one of the most developed biomarkers of HPV+ disease is circulating tumor HPV (ctHPV)^130,131. In a prospective biomarker trial of 115 patients, 2 consecutively positive ctHPV tests were associate with a positive predictive value (PPV) of 94% for recurrence following definitive chemoradiation¹³¹. Additional trials evaluating ctHPV are currently underway.

Conclusions

As our ability to interrogate genomic, transcriptomic and proteomic features of malignancies has vastly improved over the past decade, a plethora of data has emerged demonstrating the prognostic and predictive utility of molecular biomarkers. Early work in this space has been applied to targeted systemic therapy and has had a profound impact on precision oncology. Although application of histology specific and molecular signature biomarkers to guide radiation therapy is in its infancy, several promising biomarkers have been developed which may soon guide the use of radiation as well as RT dosing and/or fractionation. Much future work is still needed to prospectively validate these signatures using integral biomarker designed clinical trials before they are ready for routine clinical use.

Table 1.

Select promising biomarkers to guide radiation therapy

Disease Site	Biomarker	Clinical Utility
Breast (DCIS)	DCISionRT	DCISionRT score >2.8 is predictive for benefit of adjuvant RT to reduce risk of ipsilateral invasive recurrence24,26.
Breast (early stage)	PAM50	The LUMINA trial demonstrated a low 5-yr local recurrence risk (2.3%) of Luminal A tumors treated with BCS and adjuvant ET without adjuvant RT41.
Breast (early stage)	Oncotype Dx RS	1) Adjuvant radiation demonstrated an improvement in 5-yr OS in patients with intermediate or high RS but not low RS and is being further evaluated in NRG BR00744. 2) RS is significantly associated with locoregional recurrence risk and should be evaluated for incorporation of RNI for high-risk patients45,46.
Breast (early stage)	ARTIC	Patients with low ARTIC score demonstrated a significant 10-yr LRR benefit from adjuvant RT where as those with high score did not53.
Prostate Cancer (very low-favorable intermediate risk)	Decipher GC	GC ≤ 0.2 was associated with a high NPV for adverse pathologic features and may aid in selection of candidates for active surveillance62.
Prostate Cancer (post-prostatectomy)	PORTOS	Adjuvant radiation significantly reduced the risk of distant metastasis in patients with high but not low PORTOS score80.
Prostate Cancer (post-prostatectomy)	Decipher GC	Patients with intermediate/high GC demonstrated a significantly greater benefit in OS with the addition of ADT to salvage RT compared to those with low GC88.
Medulloblastoma (average-risk)	WNT	Wnt subgroup medulloblastoma is associated with improved event-free survival and is being evaluated for low-dose CSI100.
Oropharyngeal cancer (intact)	HPV	HPV positive oropharyngeal SCC is associated with significant improved outcomes and increased radiation sensitivity. NRG HN005 is currently evaluating dose reduced radiation with either cisplatin or nivolumab compared to standard of care.
Oropharyngeal cancer (post-op)	HPV	ECOG 3311 demonstrated no significant difference in 2-yr PFS between standard and dose reduced adjuvant RT in patients with intermediate-risk features117.