Review of emerging biomarkers in head and neck squamous cell carcinoma in the era of immunotherapy and targeted therapy

Biomarkers in head and neck squamous cell carcinoma (HNSCC) emerge rapidly in recent years, especially for new targeted therapies and immunotherapies.


| INTRODUCTION
Head and neck squamous cell carcinoma (HNSCCC) is a heterogeneous disease characterized by malignant and uncontrolled growth of cells in various sites within the head and neck areas, such as the oral cavity, larynx, oropharynx, hypopharynx, paranasal sinuses, and nasal cavity. For the purposes of this review, discussion on nonsquamous cell cancers originating in the head and neck, including nasopharyngeal carcinoma, differentiated or undifferentiated thyroid cancer, and salivary gland cancer, are outside the scope of this work. While these cancers are conventionally defined as members of the family of head and neck cancers, they are generally thought of as different entities from HNSCC. The main reasons for this distinction are due to their different behaviors with regard to tumor development and progression, patterns of relapse, sensitivity to chemotherapy and radiotherapy, and patient outcomes.
According to the National Cancer Institute (NCI), a biomarker is defined as "a biological molecule found in blood, other body fluids, or tissues, that is a sign of a normal or abnormal process, or a condition or disease. A biomarker may be used to see how well the body responds to a treatment for a disease or condition." 1 In the field of cancer treatment, biomarkers have at least 4 key roles in clinical application, including (a) assisting in the diagnosis of cancer; (b) indicating likely clinical outcomes (prognostic role); (c) aiding in patient selection for a specific treatment, based on which patients are most likely to respond (predictive role), and (d) deciding at what dosage the drug might be most effective (pharmacodynamic role). 2 The latter 3 roles are generally applicable for newer chemotherapies, targeted therapies, and immunotherapy, including antiprogrammed death (PD)-1 or anti-PD-ligand 1 (L1) agents. While the use of many biomarkers is not yet routinely available in current clinical practice, biomarkers can provide critically useful and cost-effective information. 3 In this review, we highlight the clinical evidence and utility of established HNSCC biomarkers in clinical practice, along with several emerging biomarkers under development. Note that cell line investigations were not included in this review.

| OPPORTUNITIES AND CHALLENGES IN BIOMARKER DEVELOPMENT
Before a biomarker can be adopted into routine practice to aid in clinical decision making, a series of strict processes must be undertaken during bench-to-bedside development 4,5 (Figure 1). First, a biomarker target is identified (discovery phase) and preliminarily confirmed through larger-scale, repeated laboratory experiments (confirmation phase). In subsequent clinical trials (validation and refinement phase), researchers attempt to set appropriate endpoints to validate preclinical findings in an independent patient cohort, before routine use of the biomarker is adopted (adoption phase).
However, a number of challenges exist in the biomarker development process (Figure 1). 6 First, an appropriate target must be identified. In this rapidly evolving era of highthroughput omics technology, thousands of candidate molecules can be investigated easily without an a priori hypothesis. A more specific, targeted approach is a posthoc, data-driven investigational study design, rather than conventional pathway rationale-driven or hypothesis-driven designs. 7 Setting a relevant, reproducible biomarker cutoff value to guide subsequent clinical measurement and validation can also be a difficult hurdle in biomarker test planning. 7 Furthermore, results of biomarker analyses can be notoriously inconsistent, making it difficult to draw robust conclusions. 8 Some studies might show negative or discrepant results, while the same biomarker might clearly demonstrate positive associations in other studies-for example, CCND1, 9 cMET, 10 p16, [10][11][12] EGFR, 13 and ERCC1. 14 Plausible reasons for such discrepancies might include (a) small sample sizes with inadequate controls; (b) differing study F I G U R E 1 The four phases of biomarker development and their challenges. Biomarkers are generally identified from a broad experimental data (discovery phase), then investigated and confirmed through preclinical experiments, and clinical samples from cancer patients (confirmation phase). Further investigation would generally focus on the established relevance to a specific population. Reference cutoffs for biomarkers are defined with relevance to specific clinical endpoints, and the clinical utility of the marker is then explored and validated in a homogenous patient cohort. Finally, strongly validated biomarkers are adopted into routine clinical practice and become standard testing for diagnosis, patient selection, or outcome prediction [Color figure can be viewed at wileyonlinelibrary.com] populations with true clinical variability; (c) differing treatment modalities; (d) variations in the biomarker assay, for example, different technological platforms used for detection and measurement; (e) differences in the biomarker source, for example, tissue vs liquid biopsy, or fresh vs fixed tissue; (f) varied antibody specificities and binding affinities among different batches or vendors; (g) biomarker instability, 6 with a risk of false positivity or false negativity; (h) differing statistical testing methods; and (i) other methodological differences between studies, for example, evaluation of mRNA vs protein expression. 6,8,15 It is important that test protocols mitigate the effects of such confounders in biomarker testing. For example, the use of standardized materials and methods should be used where possible. If the sample size is inadequate, investigators may need to apply cross-validation-based methods. Preselection of target populations can also be of key importance in biomarker study success, as nonpreselected populations could lead to a high trial cost and risk of failure. 16 Finally, many factors can impact the adoption of a biomarker among clinicians, including the high cost of routine testing, low power of evidence, accessibility of information, and lack of avenues for clinical feedback with regard to biomarker-directed therapies. 17 Unfortunately, the high number of challenges around biomarker development means that very few markers (as low as 0.1%) achieve a substantial clinical role, limiting the use of biomarkers in routine practice. 3,16,18

| p16: an important prognostic and predictive tissue biomarker in HNSCC
Of the many biomarkers with a prognostic role in HNSCC, the most well established and validated is p16, which plays an essential role in HNSCC. 10,80 Approximately one third of HNSCCs express p16. p16 is a widely used clinical biomarker for HPV-a well-known cause of HNSCC-and HPV status, in turn, is an important prognostic marker used for patient stratification in HNSCC. [95][96][97] Notably, in the most recent American Joint Committee on Cancer (AJCC) cancer staging book (2017), 75 a distinct staging system for HNSCC patients with positive p16 expression was recommended. The p16 protein is generally evaluated by immunohistochemistry (IHC), and HPV infection by DNA/mRNA polymerase chain reaction (PCR). HPV has proved to be a significant diagnostic and prognostic biomarker in particularly in the oropharynx 39,98 and cancers with unknown primary (CUP) presenting with neck node squamous cell carcinoma. 50,99 p16 as a surrogate for HPV demonstrated strong prognostic value in patients with oropharyngeal squamous cell carcinoma (OPSCC) in a phase III registration trial evaluating radiotherapy alone or in combination with cetuximab. 41 One study has suggested that p16 status is an important prognosticator in both OPSCC and non-OPSCC, and that the p16 positive/HPV16 negative group is likely a distinct and important subgroup for future trials. 100,101 p16 also appears to be an important prognostic marker in cutaneous HNSCC, which commonly presents as cervical metastases secondary to CUP. High p16 expression has been shown to be indicative of primary HNSCC with better survivals 10 ; however, p16 expression was not associated with improved survival in this specific subgroup of HNSCC. 10 In theory, p16 positivity would not be exclusively associated with HPV infection, considering p53 and retinoblastoma (RB) gene involvement. 102,103 The virus contains two oncogenes, E6 and E7, which, when expressed, inactivate p53 and RB, respectively. p53 is frequently inactivated in HNSCC, and dysregulation of the RB gene is increased by the expression of p16. 97 When testing for p16 and HPV, the two tests are not always concordance with each other (around 81.0%-94.6%). 39,82,104-107 It has been suggested that performing combination testing of p16 and HPV status together may improve prognostic accuracy. 108,109 p16 also has predictive value with regard to HNSCC outcomes with various specific treatments. 57,62 Patients positive for p16 have shown a greater response to, and improved overall outcomes with radiotherapy, 61 as well as improved outcomes from CTC count during treatment (CCRT), 62 vs those with p16 nonexpression. In addition, some studies indicate a predictive role for p16 status in patients receiving EGFR-targeted therapy for HNSCC, including cetuximab plus chemotherapy, 46 panitumumab plus chemotherapy (vs chemotherapy alone 57 ), and afatinib. 58,59 As excellent outcomes, in general, can be achieved in HPV-positive HNSCC, the reasonable next step is to deintensify therapy, especially radiotherapy, to minimize treatment-related toxicities and improve quality of life without compromising survival. 110,111 One thing to be noticed is that the de-intensification needs to happen carefully and only within the confines of a clinical trial. Given the distinctive pathobiology of HPV-positive HNSCC, innovative approaches targeting viral oncogenes and the immune system, integrated with the use of both established and novel biomarkers, are warranted. 3

| Predictive tissue markers for platinumbased therapy
As early as 2000, the use of cisplatin or carboplatin in combination with radiotherapy has become a standard treatment approach for HNSCC. 110,112 Several tissue biomarkers have shown a potential role in predicting response to platinum-based therapy. For example, ERCC1 protein expression may be a valuable marker for platinum chemoresistance, 14,30,32,54 106 radiotherapy toxicity, 52 and response 113 in HNSCC. However, one study in 48 HNSCC patients 68 showed no apparent role for mRNA amplification of the ERCC1 gene in terms of predicting platinum resistance, suggesting that such resistance might be attributable to non-ERCC1 pathways. 114 This may be a unique phenomenon in HNSCC; in a prospective phase III trial in nonsmall lung cancer, ERCC1 mRNA expression was able to predict acquired resistance to platinum treatment. 115 Bcl-2 is another example; evidence suggests that this marker may contribute to distant failure in HNSCC patients receiving platinum-based CCRT. 19,51 In addition, GSTsthat appear to play an essential role in the cell's defense against toxic substances-may predict platinum resistance in HNSCC. 54 MRP2 protein expression and RB protein expression were also found to be independently associated with reduced local control in patients who received CCRT. 55 Another study evaluated stromal cell-derived factor 1 (SDF-1)/CXCR4 and demonstrated predictive ability with respect to locoregional control and survival in 141 patients who underwent surgery and adjuvant CCRT. 24 It should be noted, however, that some of these studies failed to discriminate between resistance to RT vs chemoresistance.

| Predictive tissue markers for EGFRtargeted therapy
The family of human epidermal growth factor receptor (HER/EGFR/ErbB) contains 4 subtypes of EGFR members (ErbB1 to 4/EGFR, HER2-4), which play essential roles in cancer cell proliferation, vessel angiogenesis, and dissemination through downstream oncogenic signaling pathways. EGFR is overexpressed in more than 90% of HNSCC, but the loci of mutations are not in common hotspots. 49,116,117 Currently, there are two main types of EGFR-inhibition-mediated therapeutic agents, including (i) EGFR monoclonal antibodies (mAbs, that is, cetuximab and panitumumab) that target extracellular ligand binding domains; and (ii) EGFR-tyrosine kinase inhibitors (TKIs, that is, gefitinib and afatinib) that target intracellular ATP-binding pockets in tyrosine kinase domains. 118,119 In 2008, addition of cetuximab to platinumfluorouracil chemotherapy significantly improved OS compared with platinum-based chemotherapy alone as first-line treatment in patients with recurrent or metastatic HNSCC. 57,120 Cetuximab alone has also shown efficacy in platinum-refractory cases th HNSCC. 54 Interest in establishing accurate predictive markers of response to EGFR inhibitors continues to grow.
Increased EGFR gene copy number appears to be largely restricted to p16 INK4A -negative oropharyngeal cancer. 38 Biomarker studies evaluating the role of EGFR protein expression have shown inconsistent results. 38 For example, as briefly described above, p16 status appears to influence response to EGFR-targeted therapies. Patients with p16-positive tumors responded well to cetuximabbased therapy, 41,46 those with p16-negative tumors responded better to panitumumab-based therapy 57,74 and afatinib. 58,59 In the phase III LUX-Head&Neck 1 (LUX-H&N1) trial, 2nd line afatinib significantly improved PFS vs methotrexate in patients with recurrent/metastatic HNSCC. In subgroup analysis, patients who have benefited from afatinib were identified in those with p16 neg , EGFR amplified , HER3 low , PTEN high status. 59 That indicates that p16, HER3, and PTEN might serve as predictive markers in afatinib treatment. Also, in another study, high heregulin mRNA and high HER3 protein levels independently correlated with poor OS in oropharyngeal cancer patients, which indicates targeting HER3 as one of the potential treatment targets. 121 It is reasonable to assume that biomarker signatures made up of combinations of established markers such as EGFR, RB, p53, CDK2, p16, p21, and HPV E6/E7 levels, may offer a more feasible approach to response prediction in HPV-positive HNSCC. Table 2 summarizes the available evidence for prognostic and predictive biomarkers in the era of immunotherapy, focusing on PD-1 and PD-L1 inhibitors. PD-1 and PD-L1 expression currently remain the most significant tissue biomarkers. PD-L1 expression has been associated with postchemotherapy (docetaxel/platinum/5-fluorouracil regimen) status, 127 co-occurrence with p16 INK4 expression, 131 and poorer OS 124,126,128 (but improved RFS 125,129 and OS 125,130 ). Among these studies, associations between high PD-L1 expression and favorable OS were all demonstrated in post-surgery HNSCC patients. 129 It is possible that PD-L1 expression may differentially impact resectable and unresectable patients, which requires further investigation. In patients with HNSCC who underwent pulmonary metastasectomy, higher PD-L1 expression predicted poorer outcomes after palliative surgery. 134 Regarding studies of predictive markers for newer immunotherapies, PD-L1 expression was associated with a higher ORR 138 and longer OS 138 after nivolumab therapy, a favorable response to radiation (radiosensitivity), 132 and an improved response to durvalumab. 122 In addition to PD-L1 expression, microsatellite instability (MSI) predicted response to PD-L1 inhibitors in HNSCC. 139 HPV status was predictive of improved response to durvalumab. 122 Furthermore, a higher number of some subtypes of tumor-infiltrating lymphocytes (TILs), such as PD-1 + TIM-3 + CD8 + TILs and PD-1 + LAG-3 + CD8 + TILs, and higher tumor mutation burden (TMB) and CD8 + TILs, all predicted improved response to anti-PD-1 or anti-PD-L1 therapies. 133 Several recent conference presentations have also highlighted novel data regarding predictive biomarkers in the era of immunotherapy (Table 2). For example, PD-1 + CD8 + effector T cells and PD-1 + Treg cells in tumor tissue predicted response to nivolumab, 136 whereas mutational load and IFN-γ gene expression profile (GEP) predicted response to pembrolizumab. 137 However, data on these and other emerging predictors of immunotherapy response remain inconclusive. An ongoing prospective trial, PRECISION-01 (NCT03917537; www.ClinicalTrials.gov), aims to resolve some of this uncertainty by evaluating biomarker signatures in cancer tissue via whole-genome/exome sequencing, in patients with platinum-refractory HNSCC receiving nivolumab monotherapy. 140 Clinical application uses current evidence of validated cutoff values of PD-L1 expression (Table 3). In brief, each clone of PD-L1 for an individual immunotherapy drug has its validated cutoffs, although these may vary between studies. To more fully elucidate the predictive role of PD-L1 expression, a study comparing HNSCC patient populations identified by different PD-L1 assays is now underway. 143 However, PD-L1 expression on cancer tissue currently still has no bearing on the management of patients with HNSCC.

| IMAGING BIOMARKERS IN HNSCC
In recent years, advances in diagnostic imaging technologies have not only aided cancer diagnosis and staging, but have also become an important tool in predicting disease outcomes, relapse patterns, and treatment outcomes. The most commonly utilized tool is functional magnetic resonance imaging (MRI), using diffusion-weighted imaging (DWI), blood oxygen level-dependent (BOLD), and dynamic contrast-enhanced (DCE) sequences. 144 Some studies have established multidimensional prognostic or predictive signatures by combining more than one type of biomarker. For example, one or more imaging parameters (eg, maximal standard uptake value, 51 metabolic tumor volume, 145 apparent diffusion coefficient, 146 or gross tumor volume 51 ) may be combined with tissue protein expression markers, or liquid biopsy markers. 147

| BIOMARKERS FROM LIQUID BIOPSIES
Some biomarkers can be isolated from body fluids, such as blood, urine, and saliva, and objectively measured. [148][149][150][151] Important benefits of liquid biomarkers include their noninvasive nature, and convenience in terms of taking serial measurements. 152 However, a number of factors have the potential to impact the dynamics and reliability of liquid markers, limiting their clinical utility. These may include the timing of collection, 153 fluid viscosity, nutritional factors, inflammatory conditions, secretory gland injury, or other environmental factors. Table 4 summarizes some examples of liquid biopsy markers in HNSCC; however, it should be noted that some data lack proper validation.
Blood is a relatively stable body fluid; several serum protein markers have traditionally been used to help predict outcomes in HNSCC patients, including CDK4, 154 midkines, 178 tumor necrosis factor (TNF)-like weak inducer of apoptosis (TWEAK), 187 IL-2R, 186 and VEGF-A. 188 However, some are not always cancer-specific, giving rise to problems in distinguishing between malignant and inflammatory disease. [189][190][191][192] More recently, markers from liquid biopsies that are directly related to cancer cells or cancer-produced molecules have been employed, including circulating tumor cells (CTCs), 147,[150][151][152][153][154][155][156][157][158][159][160][161][162][163][164]193,194 cell-free DNA (cfDNA)/ circulating tumor DNA (ctDNA), [187][188][189][190][191][192]195,196 cfRNA, 184 and exosomes. [196][197][198] It is generally accepted that CTCs may play an important role in cancer metastasis. 199 Data suggest that an elevated baseline CTC count is associated with advanced stage of HNSCC, 153,161 risk of relpase, 161 and a poor prognosis, [158][159][160]162,[166][167][168][169][170]172 while a declining CCRT could indicate an improved prognosis and treatment response. 164,172 In some patients, an increased CTC count has been observed after HNSCC treatment by surgery 153 and radiotherapy, 173 which is thought to possibly reflect stimulation of tumor cell dissemination and a poorer prognosis; although, the exact reasons for this observation remain unclear. A subgroup of CTCs express podoplanin, a known prognostic factor for HNSCC. Evidence suggests that an elevated podoplanin:EpCAM ratio in CTCs may be associated with a poor prognosis and treatment failure. 165 In addition, PD-L1-positive CTCs may play a role in predicting response to immunotherapy. 163,182 With regard to cfDNA, mitochondrial cfDNA content appears to be strongly associated with HNSCC in patients with lifestyle risk factors for the disease. 155 Various other liquid biomarkers may provide additional diagnostic and prognostic information in HNSCC. Examples include elevated serum miR-21, 179   exosomes, 181 and presence of the TGFβ 1 genotype, 12 all of which may be associated with a poorer prognosis. Several important advances have been achieved in recent years, that have improved the potential of liquid biopsy markers as robust clinical tools. Such advances include (a) estimation of tumor mutational burden through evaluation of plasma cfDNA 200 ; (b) more therapy-directed applications of CTCs, for example, PD-L1 expression in CTCs 163,165,182 ; (c) use of CTCs as a longitudinal monitoring tool to detect minimal residual disease after curative surgery 171 ; (d) CTC-derived xenografts used as surrogates to study tumor biology [201][202][203] ; and (e) three-dimensional CTC cultures 204 and CTC-derived organoids to aid in individualized precision medicine. 205

| FUTURE PERSPECTIVES
The rapid evolution of the background understanding of cancer physiology and biology has affected every aspect of disease management and patient care. In this evolving era of precision medicine, there is an ever-building need, including in HNSCC, for novel prognostic and predictive biomarkers with robust clinical application.
Ideally, biomarker trials should be designed based on an actual clinical need; as such, peer review panels evaluating biomarker research proposals now pay close attention to the potential clinical utility of biomarker tests. 15 Given the challenges of biomarker development discussed above, concerted efforts also need to be made to harmonize assays, methodologies, and cutoffs, to ensure consistency of results and allow accurate extrapolation of trial data to the clinical setting.
Identification of driver mutations relevant to specific targeted therapies remains an ongoing area of research. A number of genetic and histological markers under development may prove integral to patient selection for the testing of novel targeted therapies-for example, those being evaluated in the ongoing National Cancer Institute Molecular Analysis for Therapy CHoice (NCI-MATCH) Precision Medicine Clinical Trial. It is hoped that more purposeful patient selection will enable inhibition of specific aspects of oncogenic pathways, and optimize the applicability of trial data, with the goal of stabilizing disease and improving survival in the greatest number of patients possible. 18 Finally, future development of artificial intelligence technology may help predict clinical outcomes more precisely than current technology and traditional statistical analysis. 206