Deciphering site‑specific histopathological parameters with potential clinical value in head and neck squamous cell carcinomas

Introduction

Head and neck carcinoma accounts for 4.6% of all cancers worldwide, and there has been an annual 1.4-fold increase in the number of new cases per year during the last decade (1,2). Squamous cell carcinomas (SCC) comprise the most common type of malignancy in this wide anatomical region, which includes the oral cavity, pharynx, and larynx (3). Although tobacco smoke and alcohol consumption are traditionally considered major risk factors for head and neck SCC (HNSCC), viruses are also accountable (4). Two types of viruses, namely, Human Papilloma Virus (HPV) and Epstein-Barr Virus, are strongly associated with oropharyngeal and nasopharyngeal SCC, respectively, since virus-associated carcinomas exhibit a predilection for sites with lymphoid-based mucosa (3,5). In fact, according to the latest recommendations of The College of American Pathologists, all newly diagnosed oropharyngeal SCC should be examined by immunohistochemistry, using p16 as a surrogate marker for potential HPV infection (6), while HPV-related oropharyngeal cancer is currently considered as a unique disease entity in both the pathological (7) and clinical (8) context. Emerging molecular drivers such as lncRNA BLACAT1 have been implicated in adjacent sites like hypopharyngeal SCC (9), underscoring the need to integrate morphologic and transcriptomic data in future studies. Epigenetic dysregulation is a common feature across cancers; chromatin-remodeling complex vulnerabilities may provide novel therapeutic targets in HNSCC, as recently suggested for a different type of tumors (10).

Several studies have reported the significance of both tumor cells and tumor microenvironment histopathological characteristics in HNSCC. Emerging data have highlighted the importance of phenotypic heterogeneity in these carcinomas and its role in the development and progression of the tumors as well as in treatment efficacy (11). In this context, the prognostic value of traditional grading is debatable, whereas other histopathological features, i.e. worst pattern of invasion and intensity of lymphocytic infiltration in the invasive margin of HNSCC, seem directly associated with patient overall survival (OS) (12,13).

In the present study, we assessed histopathological parameters of tumor cells and their microenvironment in tumor compartments (center and invasive margin) of HNSCC originating in different anatomical sites. We evaluated these parameters against each other and against clinicopathological characteristics of the patients to identify potential risk factors with clinical relevance.

Materials and methods

Tissue samples and patient data

Formalin-fixed paraffin-embedded tissue samples from patients with histologically confirmed HNSCC and available clinical, histopathological, treatment, and outcome data were retrieved from the Hellenic Cooperative Oncology Group (HeCOG) clinical database and biologic material repository. Since nasopharyngeal carcinoma is generally considered a biologically different disease compared to SCC of the rest of the head and neck region, especially in Far East and Mediterranean basin, we explored this neoplasm separately on genomic, histopathological and clinical basis (14–16) and it was excluded from this study. In the same line, p16 positive HNSCC considered as HPV-related, which are biologically different from other HNSCC (7), were excluded, as well. Patients with HNSCC were diagnosed and treated between 1999 and 2018. Written informed consent was obtained from all patients for the use of their data and biologic material for research purposes. The protocol was approved by the Bioethics Committee of the Aristotle University of Thessaloniki, School of Health Sciences, Faculty of Medicine (approval no. 5/18.12.2019; Thessaloniki, Greece).

New Hematoxylin and Eosin (H&E) sections from all tissue blocks were reviewed for histological evaluation and tumor tissue adequacy (tumor cells and stroma). The tumor surface area was separated into tumor center and invasive margin according to the literature (17,18). As such, the invasive margin was defined as the region around the border where normal tissue meets malignant cells, extending by 1 millimeter. Samples from advanced HNSCC that were not obtained from the primary site (e.g., metastatic lymph nodes) or where a primary site could not be determined, were annotated as ‘local spread’. Based on the availability of clinical data and adequate tumor tissue, 248 tumors from the same number of patients were examined. Patient age at first diagnosis ranged from 20.9 to 85.1 years, with a median of 62 years.

Histological evaluation

Parameters assessed on whole H&E sections were: i) Keratinization (present/absent). Tumors exhibiting squamous maturation features on less than 10% of their surface were not classified as keratinizing (19); ii) area occupied by keratin pearls as percentage of the entire tumor area; iii) presence of specific spatial distribution of neutrophils, that is, neutrophils in tumor necrotic areas and in close proximity to dyskeratotic cells, as well as microabscesses in keratinization; iv) histologic grading, according to the three-tier grading system (G1, G2, G3) (20); v) extent of tumor necrosis, as absent, focal (≤10% of the tumor area), moderate (11–29% of the tumor area), or extensive (≥30% of the tumor area) (21); and the presence of necrosis in keratinization areas. However, since only one case exhibited extensive necrosis, it was included in the moderate group; vi) anaplastic cells defined as cells with lack of differentiation, showing pleomorphism, nuclear abnormalities, atypical mitoses and loss of polarity (present/absent) (22); vii) cell cannibalism, which corresponds to cell-in-cell phenomenon, a process of non-apoptotic cell death, where one cancer cell surrounds another cancer cell (present/absent) (23); viii) tumor stroma classification, as myxoid, when there was an amorphous stromal substance comprising of slightly basophilic extracellular matrix; as keloid-like, regulated by specific channels (24), when thick bundles of hypocellular collagen with bright eosinophilic hyalinization were evident; or as fibroblastic, when only fine mature collagen fibers were evident (25); ix) tumor infiltrating lymphocyte (TIL) density; according to international guidelines (17,26), ten high-power fields of each section excluding pre-existing lymphoid background areas (17,27), were assessed, and the mean average was estimated; and x) perineural and lymphovascular invasion (present/absent). The presence of calcification and giant cell reaction was also recorded.

Eosinophil number per mm2 was evaluated similarly to CD8+ counts on Tissue MicroArray (TMA) sections (28).

Histological parameters in the invasive margin were examined separately, based on previous studies (5,12,13,29): keratinization was assessed as absent (0.5% of the invasive margin area), low (6–20%), moderate (21–50%), and high (<50%); cellular atypia and pleomorphism were estimated as mild, moderate, and severe; mature squamous cells with angular shape, eosinophilic cytoplasm, and intercellular bridges were recorded in a four-tier scale, based on their abundance (0–25%, 26–50%, 51–75%, and <75%); lymphocytic host response along the invasive margin was recorded as dense (dense lymphocytic response rimming the tumor and/or lymphoid nodules at advancing edge in each 4× field), intermediate (intermediate lymphocytic response and/or lymphoid nodules in some but not all 4× fields), and weak (sparse lymphocytic response without lymphoid nodules) (13); and perineural invasion was assessed separately in this tumor compartment. The worst pattern of invasion, defined as the least differentiated part of the invasive margin of tumors, included five categories, namely, pushing border, finger-like growth, large tumor islands (>15 cells/island), small tumor islands (<15 cells/island), and tumor satellites ≥1 mm away from the main tumor.

Immunohistochemistry

Immunohistochemistry was performed on 3-μm TMA sections. To confirm HPV-negative status, p16INK4A mouse monoclonal antibody (clone IHC116, GenomeMe, Lab Inc., Richmond, BC, Canada) was applied at 1:50 dilution, on a Bond Max Autostainer using BOND Polymer Detection kit. Antigen retrieval was performed with EDTA (pH 9) for 40 min, followed by primary antibody (p16INK4A) incubation at 37°C for 30 min. For immune profiling, CD8 mouse monoclonal antibody (clone C8/144B, Dako, Glostrup, Denmark) was used at 1:60 dilution. For antigen retrieval, EDTA (pH 9) was used for 20 min, before primary antibody incubation at 37°C for 30 min. Staining was performed on a Dako autostainer using the EnVision™ FLEX+ (Agilent, Santa Clara, CA) visualization system. PD-L1 was assessed with a mouse monoclonal antibody (clone 22C3, Dako, Glostrup, Denmark) at 1:100 dilution on a Dako autostainer by using the EnVision™ FLEX+ visualization system (Agilent, Santa Clara, CA). For antigen retrieval, EDTA (pH 9) was performed for 30 min, followed by primary antibody incubation at 37°C for 30 min. All slides were counterstained with hematoxylin. Both positive and negative controls were simultaneously stained.

In regard to immunohistochemical evaluation, PD-L1 expression was assessed in both tumor and immune cells, according to the guidelines (30). For tumor cells, PD-L1 positivity was defined as complete and/or partial circumferential linear cellular membrane staining of any intensity, as assessed by the tumor proportion score (TPS). Immune cells were evaluated as the proportion of the tumor area occupied by any discernible PD-L1 staining of any intensity and the combined positive score (CPS) was estimated in each case (30). CPS ≥1 was considered positive. Staining interpretation was performed by two experienced pathologists (SET, TK). CD8+ cell counts were obtained from the entire area of each 1.5 mm diameter core. The density of positive cells was assessed as the ratio of CD8+ cells per mm2 of core surface area. Average values were used for tumors represented by multiple cores. Spatial distribution of CD8+ T cells was assessed by comparing two tissue cores of each tumor to determine whether cells were evenly distributed, clustered or concentrated in specific regions or showed different values between the cores (28). For p16, strong and diffuse nuclear and/or cytoplasmic positivity in ≥70% of neoplastic cells was used as a cut-off according to standard guidelines for HNSCC, with a positive result indicating HPV-related HNSCC (31). As aforementioned, p16 immunohistochemistry (Fig. S1) was applied to all available HNSCC cases and HPV-related carcinomas were excluded from the present study; accordingly, all tumors analyzed here were p16 negative.

Statistical analysis

Histological parameters, categorical and continuous, were evaluated against clinicopathological patient and tumor characteristics and against each other. In case of two categorical variables χ2 and Fisher's tests were applied. In case of comparing a continuous variable across different groups, the Kruskal-Wallis(if more than two groups) and the Mann-Whitney (if two groups) tests were used. In case of post hoc comparisons, Kruskal Wallis test, followed by Mann-Whitney U test and Bonferroni correction was applied.

For the assessment of risk factors for OS, the log-rank test and Cox regression were applied. For multivariable analyses, we selected risk factors with a P-value lower than 0.2, which were further filtered by using Akaike's information criterion for entry into the multivariable model.

The significance level for univariable analyses was set at 5%, with P<0.05 considered to indicate a statistically significant difference. All analyses were conducted in R (v. 4.3.1; http://www.R.project.org/).

Results

Patient and tumor characteristics

Most patients were men, heavy smokers, and diagnosed with stage III or IV disease. The majority underwent surgical treatment, while 26.25% received various therapeutic modalities in the first line setting. Based on tumor location as per histopathology report, SCC were categorized into oral, oropharyngeal, hypopharyngeal, and laryngeal, the two latter categories being analyzed as one. Twenty-seven cases of local spread specimens were incorporated in the study. Most tumor tissues were obtained at disease diagnosis and were surgical specimens. Seventy-nine of them included the invasive margin of the tumor. The resection margins were free of carcinoma/high grade dysplasia. All available clinicopathological data are presented in Table I.

Table I. Head and neck squamous cell carcinoma patient demographics and clinicopathological data.

Table I.

Head and neck squamous cell carcinoma patient demographics and clinicopathological data.

Characteristic	Value
Median age at first diagnosis, years	62 (54.23–70.1;
(IQR min-max; range)	20.9–85.1)
Sex (n=248)
Female	31 (12.5)
Male	217 (87.5)
Smoking packs (n=225)
No	25 (11.1)
1-10 pack years	9 (4.0)
11-20 pack years	27 (12.0)
>20 pack years	164 (72.9)
Alcohol abuse (n=226)
No	68 (30.1)
Mild	53 (23.5)
Moderate	41 (18.1)
Heavy	64 (28.3)
Disease stage (n=229)a
I	25 (10.9)
II	27 (11.8)
III	65 (28.4)
IV	112 (48.9)
Adjuvant treatment (n=242)
No	210 (86.8)
Chemo	1 (0.4)
RT	31 (12.8)
First line treatment (n=240)
No	177 (73.8)
Chemo	38 (15.8)
CT-RT	7 (2.9)
RT	18 (7.5)
Death (n=246)
No	59 (24.0)
Yes	187 (76.0)
Tumor site (n=248)
Oral	35 (14.1)
Oropharyngeal	30 (12.1)
Laryngohypopharyngeal	156 (62.9)
Local spreadb	27 (10.9)
Sample status (n=242)
First diagnosis	202 (83.5)
Pretreatedc	23 (9.5)
Relapse, naïved	17 (7.0)
Type of specimen (n=248)
Biopsy	78 (31.5)
Surgical	170 (68.5)

{ label (or @symbol) needed for fn[@id='tfn1-ol-30-6-15315'] } Data are presented as n (%), unless otherwise indicated.

a Stage was pooled into the main 4 categories according to AJCC 8th ed;

b tumor spread to adjacent anatomical structures and/or lymph nodes;

c radiotherapy +/- chemotherapy;

d preceding surgical treatment only. RT, radiotherapy; chemo, chemotherapy; CT-RT, chemotherapy and radiotherapy; IQR, interquartile range.

Phenotype characteristics

Detailed histopathological and immunophenotypic characteristics of the examined HNSCC are shown in Table II.

Table II. Histopathological and immunohistochemical characteristics of head and neck squamous cell carcinoma cases.

Table II.

Histopathological and immunohistochemical characteristics of head and neck squamous cell carcinoma cases.

Characteristic	Value
Grade (n=248)
G1	42 (16.9)
G2	155 (62.5)
G3	51 (20.6)
Grade heterogeneity (n=248)	24 (9.7)
Tumor variant (n=248)
Keratinizing	196 (79.0)
Non keratinizing	52 (21.0)
Anaplastic features (n=248)	138 (55.6)
Cell cannibalism (n=248)	139 (56.0)
Desmoplastic stromal reaction (n=222)
Fibroblastic	134 (60.4)
Keloid-like	27 (12.1)
Loose/myxoid	61 (27.5)
Stromal reaction heterogeneity (n=222)	37 (16.7)
Necrosisa (n=248)
Absent	220 (88.7)
Focal	21 (8.5)
Moderate-extensive	7 (2.8)
Microabscesses (n=248)	13 (5.2)
Necrosis in keratinization (n=248)	68 (27.4)
Neutrophils in necrosis (n=248)	14 (5.6)
Neutrophils in dyskeratosis (n=248)	126 (50.8)
Perineural invasionb (n=248)	12 (4.8)
Lymphovascular invasion (n=248)	4 (1.6)
CD8 heterogeneity (n=159)	41 (25.8)
PD-L1 status (n=215)
<1	190 (88.4)
1-20	22 (10.2)
≥20	3 (1.4)
PD-L1 heterogeneity (n=215)	12 (5.6)
Lymphocytic host response in IM (n=79)
Weak	28 (35.4)
Intermediate	27 (34.2)
Dense	24 (30.4)
Perineural invasion in IM (n=79)	9 (11.4)
Multinucleated macrophages (n=248)	29 (11.7)
Median keratin pearlsa (n=248) [IQR]	0 [0, 5]
Median eosinophils/mm2 (n=214) [IQR]	0.314 [0, 3.88]
Median TIL Salgado density (n=242)	20.75
[IQR]	[11.15, 32.35]
Median CD8+/mm2 (n=215) [IQR]	198.05
	[98.45, 420.37]

{ label (or @symbol) needed for fn[@id='tfn6-ol-30-6-15315'] } Data are presented as n (%), unless otherwise indicated.

a % of tumor surface;

b assessed in the entire tumor area. IM, invasive margin; IQR, interquartile range.

The most prevalent tumor variant was keratinizing of intermediate grade (G2) with anaplastic features and cell cannibalism (Fig. 1A). Among the laryngeal SCC, two were designated as basaloid SCC variants. Two additional basaloid SCC were found; one located in the oral cavity and one in the oropharynx. Neutrophils in dyskeratosis (Fig. 1B) were quite common, whereas keratin pearls in ≥5% of the tumor area, microabscesses, or necrosis in keratinization (Fig. 1C) were observed in a minority of tumors. Perineural and lymphovascular invasion were rare. Grade (Fig. 1D and E) heterogeneity was observed in a considerable number of tumors. Specifically, while most of them presented with intermediate grade (G2), ‘miniscule’ areas with features favoring either low (G1) or high (G3) grade frequently coexisted. In all cases with grading heterogeneity, the final grade was assigned based on the worst component, if it comprised more than 5%. Accordingly, stromal desmoplastic reaction presented with a variety of features, from loose myxoid to dense collagenous, even within the same tumor, in respect of stromal heterogeneity. In such cases, stroma was characterized based on the predominant pattern.

Figure 1. Histopathological features of HNSCC cases: (A) Cell-in-cell structures (arrows) in a case of laryngeal carcinoma (H&E ×400). (B) Neutrophils in dyskeratosis (encircled)/inset: neutrophils in close proximity to dyskeratotic cell in a case of laryngeal carcinoma (H&E ×400). (C) Necrosis in keratinization in a case of laryngeal carcinoma (H&E ×200); Grading heterogeneity with (D) well differentiated and (E) moderately differentiated areas within a single case of laryngeal carcinoma (H&E ×200; H&E ×200). (F) Multiple worst patterns of invasion along the invasive margin within the same case of oral carcinoma (H&E ×100). (G) Perineural invasion in the invasive margin of an oral carcinoma (H&E ×100). Heterogeneity in T8 lymphocytic density in an oropharyngeal case presenting (H) abundant CD8+ T cells in one tissue core of the tumor, and (I) only a few in the other (CD8 ×200). HNSCC, head and neck squamous cell carcinoma.

The infiltrative margin in 63 (79.7%) out of 79 evaluable tumors was characterized by the absence of keratinization. In the same tumor area, heterogeneous worst pattern of invasion was identified in 32/79 cases (40.5%) (Fig. 1F); all degrees of lymphocytic host response density were equally observed, while perineural invasion (Fig. 1G) was noticed in 11.4% of the tumors.

With respect to immune cell infiltrates, 10.3% of tumors exhibited eosinophils >14.2/mm2, while 15% of tumors had TIL density >50% and/or CD8+ >700/mm2. CD8+ infiltrates were heterogeneously present in 25.8% of the tumors (Fig. 1H and I). Only 11.6% exhibited any degree of PD-L1 positivity (Fig. S1).

Site specificity of generic histopathological characteristics

Histological characteristics significantly differed among HNSCC from different anatomical areas, namely, laryngohypopharyngeal, oropharyngeal, and oral carcinomas, as shown in Table III. Regarding some of these characteristics, the majority of laryngohypopharyngeal tumors exhibited anaplastic features and cell cannibalism. Oropharyngeal tumors featured higher grade and higher rates of tumor necrosis, while being devoid of eosinophils. In contrast, eosinophils, keratin pearls, and perineural invasion were more common in oral carcinomas. In addition, keratinization was prevalent among laryngohypopharyngeal tumors (P<0.001). Regarding the invasive margin (Table III), perineural invasion appeared more frequently in oral and oropharyngeal tumors.

Table III. Associations between clinicopathological variables and tumor site.

Table III.

Associations between clinicopathological variables and tumor site.

Parameter	Oral	Oropharyngeal	Laryngohypopharyngeal	P-value
Smoking packs, n (%)				0.009
No	6 (24%)	2 (8.3%)	17 (11.1%)
1-10	4 (16%)	2 (8.3%)	3 (2%)
11-20	4 (16%)	2 (8.3%)	20 (13.1%)
>20	11 (44%)	18 (75.1%)	113 (73.8%)
Disease stage				0.009
I	8 (28.6%)	0 (0%)	15 (9.9%)
II	4 (14.3%)	4 (16%)	19 (12.5%)
III	3 (10.7%)	6 (24%)	53 (34.9%)
IV	13 (46.4%)	15 (60%)	65 (42.7%)
Grade				0.002
G1	12 (34.3%)	4 (13.3%)	23 (14.7%)
G2	21 (60%)	15 (50%)	106 (68%)
G3	2 (5.7%)	11 (36.7%)	27 (17.3%)
Necrosis in keratinization				<0.001
Absent	34 (97.1%)	27 (90%)	97 (62.2%)
Present	1 (2.9%)	3 (10%)	59 (37.8%)
Anaplastic features				0.006
Absent	21 (60%)	17 (56.7%)	55 (35.3%)
Present	14 (40%)	13 (43.3%)	101 (64.7%)
Cell cannibalism				<0.001
Absent	25 (71.4%)	18 (60%)	52 (33.3%)
Present	10 (28.6%)	12 (40%)	104 (66.7%)
Perineural invasion (entire tumor area)				0.004
Absent	29 (82.9%)	28 (93.3%)	152 (97.4%)
Present	6 (17.1%)	2 (6.7%)	4 (2.6%)
Perineural invasion (invasive margin)				0.004
Absent	11 (64.7%)	0 (0%)	59 (96.7%)
Present	6 (35.3%)	1 (100%)	2 (3.3%)
Tumor variant and cell cannibalism status				<0.001
Keratinizing without cell cannibalism	20 (26.3%)	13 (17.1%)	43 (56.6%)
Non keratinizing without cell cannibalism	5 (26.3%)	5 (26.3%)	9 (47.4%)
Keratinizing with cell cannibalism	9 (8.8%)	7 (6.9%)	86 (84.3%)
Non keratinizing with cell cannibalism	1 (4.2%)	5 (20.8%)	18 (75%)
Median keratin pearls, % of tumor surface (IQR)	6.31 (8.64)	3.13 (6.33)	5.81 (12.92)	0.034
Median TIL density, % (IQR)	22.53 (12.39)	30.26 (20.71)	23 (16.3)	0.265
Median CD8+/mm2 (IQR)	395.63 (365.59)	305.74 (269.88)	273.58 (281.18)	0.123
Median eosinophils/mm2 (IQR)	15.99 (29.16)	2.57 (5.09)	13.51 (74.71)	0.003

[i] IQR, interquartile range.

Next, we profiled histological parameters for all tumors (Fig. 2). By definition, parameters related to keratinization were identified at significantly higher rates in keratinizing tumors (grade and necrosis in keratinization, neutrophils in dyskeratosis; all P-values <0.001). In comparison, anaplastic features, cell cannibalism and perineural invasion were almost equally distributed among tumors with and without keratinization. Modeling parameters related and not related to keratinization revealed a site-specific association; that is, concomitant keratinization and cell cannibalism were rare among oral and oropharyngeal tumors (P<0.001) (Table III). With respect to clinical parameters, laryngohypopharyngeal and oropharyngeal carcinomas were more frequent in heavy smokers (P=0.009) and in patients with stage III and IV disease (P=0.009). No further clinicopathological associations were observed.

Figure 2. Profile map of histological parameters focused on keratinization. Grade 1, green; grade 2, pink; grade 3, dark blue.

No associations with tumor site were found for worst pattern of invasion and lymphocytic host response in the invasive margin. Similarly, there were no associations for tumor variant, tumor necrosis, neutrophils in necrosis or in dyskeratosis, microabscesses, stromal reaction, lymphovascular invasion, multinucleated macrophages, CD8 and PD-L1 status.

Associations between clinicopathological parameters and survival

All parameters were examined for their association with OS at a follow-up period of 26 years (24/4/1997-30/11/2023). The results of a univariable analysis are presented in Table IV.

Table IV. Univariable analysis with respect to OS for the parameters studied in the entire cohort and in the invasive margin.

Table IV.

Univariable analysis with respect to OS for the parameters studied in the entire cohort and in the invasive margin.

A, Entire cohort
Parameter	Cases	Events	HR	95% CI	P-value
Sex					0.102
Female	30	17	1	-
Male	216	164	1.51	(0.92, 2.49)
Smoking					0.264
No	25	20	1	-
Yes	200	147	0.76	(0.47, 1.23)
Smoking pack years; only smokers			1.03	(0.88, 1.22)	0.077
Alcohol abuse					<0.001
No	68	44	1	-
Mild	53	34	0.93	(0.59, 1.46)
Moderate	41	31	1.44	(0.91, 2.28)
Heavy	64	58	2.60	(1.73, 3.89)
Disease stage					<0.001
I	25	13	1	-
II	25	17	1.01	(0.49, 2.11)
III	65	38	0.83	(0.44, 1.56)
IV	112	99	2.46	(1.38, 4.4)
Tumor site					<0.001
Oral	33	25	1.57	(1.01, 2.43)
Oropharyngeal	30	21	1.39	(0.87, 2.23)
Laryngohypopharyngeal	156	112	1	-
Local spread	27	23	2.38	(1.51, 3.75)
Tumor variant					0.616
Non keratinizing	52	35	1	-
Keratinizing	194	146	1.1	(0.76, 1.59)
Microabscesses					0.433
Absent	233	173	1	-
Present	13	8	0.75	(0.37, 1.53)
Grade					0.567
G1	41	27	1	-
G2	154	119	1.22	(0.8, 1.85)
G3	51	35	1.3	(0.78, 2.15)
Necrosis in keratinization					0.838
Absent	178	128	1	-
Present	68	53	1.03	(0.75, 1.43)
Neutrophils in dyskeratosis					0.507
Absent	120	85	1	-
Present	126	96	1.10	(0.82, 1.48)
Tumor necrosis (% of tumor surface)					0.147
Absent	218	158	1	-
Focal	21	16	1.09	(0.65, 1.83)
Moderate-extensive	7	7	2.11	(0.98, 4.55)
Neutrophils in necrosis					0.992
Absent	232	170	1	-
Present	14	11	1.00	(0.54, 1.85)
Anaplastic features					0.804
Absent	109	81	1	-
Present	137	100	0.96	(0.72, 1.29)
Cell cannibalism					0.495
Absent	108	79	1	-
Present	138	102	0.90	(0.67, 1.21)
Perineural invasion					0.095
Absent	70	47	1	-
Present	8	7	1.71	(0.9, 3.25)
Multinucleated macrophages					0.062
Absent	217	164	1	-
Present	29	17	0.62	(0.37, 1.03)
Age at 1st diagnosis	246	181	1.01	(1, 1.02)	0.127
Keratin pearls (% of tumor surface)	246	181	1.00	(0.99, 1.02)	0.459
TIL density (%)	240	177	0.99	(0.99, 1)	0.279
Eosinophils/mm2 average	212	154	1.00	(1, 1)	0.323
CD8+/ mm2 average	214	157	1.00	(1, 1)	0.184
CD8 heterogeneity					0.030
Homogeneous	118	90	1	-
Heterogeneous	40	25	0.61	(0.39, 0.96)
B, Invasive margin
Parameter	Cases	Events	HR	95% CI	P-value
Lymphocytic host response	78	54	0.56	(0.39, 0.81)	0.002
Perineural invasion present	8	7	2.52	(1.12, 5.66)	0.025

[i] Cases refers to the number of patients at the beginning of the study; events refers to the number of patients that developed the event (i.e. died) at the end of the study.

Alcohol consumption and disease stage had significant impacts on patient OS. CD8+ heterogeneity (P=0.038) was associated with a favorable prognosis, whereas perineural invasion in the invasive margin (P=0.025) appeared to be an adverse prognosticator. In addition, weak lymphocytic infiltrates in the invasive margin had a negative impact on OS (Fig. 3).

Figure 3. Significant associations of overall survival with CD8 heterogeneity in the entire tumor area as well as with PNI and LHR in the invasive margin. PNI, perineural invasion; LHR, lymphocytic host response.

A subgroup analysis was conducted by separating samples into three distinct groups based on the primary tumor site: oral cavity, oropharynx, and laryngohypopharynx. This additional analysis showed that heavy alcohol abuse and advanced disease stage were significantly associated with worse OS in laryngohypopharyngeal tumors (P<0.001). Furthermore, within the same tumor site, perineural invasion (entire cohort and invasive margin, P=0.007) and weak lymphocytic host response in the invasive margin (P=0.028) were also significantly linked to poorer OS. In oral cavity SCC, weak lymphocytic host response in the invasive margin (P=0.052) was associated with an adverse prognosis. Detailed results of these findings are presented in Table SI, Table SII, Table SIII and prognostic histological features by site are illustrated in Fig. S2.

A multivariable analysis (Table V) revealed that alcohol abuse, disease stage, and perineural invasion were independently associated with unfavorable OS. In the invasive margin, weak lymphocytic infiltrates and perineural invasion remained independent predictors for unfavorable OS. No additional associations between the clinicopathological parameters studied and OS were observed. Based on these results, we recommend histopathological parameters that should be included, among others, in the histological reports, in each case of HNSCC (Table VI).

Table V. Multivariable analysis with respect to OS for the parameters studied in the entire cohort and in the invasive margin.

Table V.

Multivariable analysis with respect to OS for the parameters studied in the entire cohort and in the invasive margin.

A, Entire cohort
Parameter	Cases	Events	HR	95% CI	P-value
Alcohol abuse	226	167	1.27	(0.95, 1.7)	0.105
Disease stage	227	168	1.67	(1.11, 2.51)	0.013
Perineural invasion present	8	7	2.81	(1.05, 7.49)	0.039
B, Invasive margin
Parameter	Cases	Events	HR	95% CI	P-value
Lymphocytic host response	80	56	0.58	(0.4, 0.83)	0.003
Perineural invasion present	8	7	2.2	(0.97, 4.98)	0.060

[i] Cases refers to the number of patients at the beginning of the study; events refers to the number of patients that developed the event (i.e. died) at the end of the study.

Table VI. Key histopathological parameters recommended for routine reporting regardless of head and neck squamous cell carcinoma anatomic site.

Table VI.

Key histopathological parameters recommended for routine reporting regardless of head and neck squamous cell carcinoma anatomic site.

Category	Parameter	Assessment/reporting
Entire tumour	Histological typea	Conventional SCC/other (specify)
	Perineural invasiona	Present/absent
	TIL density	% of all stromal mononuclear cells (including lymphocytes and plasma cells)
Invasive front	Lymphocytic host response	Weak/moderate/dense
	Perineural invasion	Present/absent
Immunohistochemistry	p16 immunohistochemistrya	Positive/negative
	PD-L1 immunohistochemistry	CPS

a Included in current RCPath and CAP core datasets for HNSCC (p16 immunohistochemistry for oropharyngeal SCC).

Discussion

Grading is one of the primary prognostic factors in malignant neoplasms; however, it does not seem to be an accurate prognosticator for HNSCC, possibly due to the heterogeneity of cytomorphological and architectural features in a pattern previously described as ‘hybrid SCC variant’ (19). Accordingly, we identified limited areas of keratinization in non-keratinizing tumors and anaplastic cells in otherwise well-differentiated carcinomas. Our observation of heterogeneity beyond neoplastic cells in microenvironmental components, such as stroma and TILs/CD8+ cells, supports the view of HNSCC as heterogeneous ecosystems.

The assessment of routinely reported histological features may assist in developing an optimal prognostic tool in line with the ongoing effort to establish multifactorial grading systems (5,32). However, we first need to identify which of the histological parameters are risk factors and whether they are site-specific. In this study, as expected, keratinizing tumors exhibited features related to keratinization, and non-keratinizing tumors were mainly characterized as high-grade. Anaplastic features, cell cannibalism, and perineural invasion were observed independent of tumor variant. Comparisons between the SCC of different anatomic locations revealed site-specific characteristics. In oral SCC, high grade was not prevalent, and a common feature was perineural invasion, as previously reported (33). Eosinophil density was higher, as well (34). Necrosis in keratinization, anaplastic cells, and cell cannibalism were mainly present in laryngohypopharyngeal SCC. Interestingly, cell cannibalism and anaplastic features were relatively common in laryngohypopharyngeal SCC, even in low-grade tumors, which is in agreement with previous reports (23). Of note, the literature contains no similar comparative morphological studies based on the HNSCC anatomical site and the observed differences could be used in developing site-specific prognostic profiling tools.

In addition, along the invasive margin, perineural invasion and weak lymphocytic host response were adverse prognosticators independently of the anatomical site. Hence, both parameters should be included in all HNSCC histopathological reporting guidelines, not only in oral carcinomas. Perineural invasion proved to be an independent prognosticator in this study, regardless of nerve diameter and intratumoral localization, in line with previous publications (35,36).

There are ongoing studies focusing on the spatial heterogeneity of the immune cells found in the tumor microenvironment with potential therapeutic implications (37–40). The observed heterogeneity of CD8+ T-cell distribution has been previously mentioned in HNSCC (41,42) without any reference to its prognostic relevance. The association of heterogeneous CD8+ infiltrates with OS described in this paper seems worth further investigation in terms of its therapeutic implications, particularly with respect to immune checkpoint inhibitors. In our series, PD-L1 evaluation showed no significant association with OS, a finding in agreement with previously published data (43). It is emphasized, however, that only a minority of our cases showed PD-L1 positivity, in contrast to former studies (44), raising concerns about its accurate evaluation on TMAs or small biopsy specimens (45).

Recent AI pipelines that fuse histology with genomic data have already achieved prognostic accuracy in other solid tumors (46) and illustrate the potential for similar multimodal models in HNSCC. Digital pathology systems leverage image analysis algorithms and machine learning to identify and quantify microscopic features with greater precision than the standard evaluation (47). Algorithms can be trained to recognize characteristic patterns of LHR-IM and PNI by analyzing H&E-stained slides and even to identify subtle changes in histology, possibly missed by pathologists (48) or to determine the clinical significance of spatial distribution of TILs (49). Machine learning models can be trained on annotated datasets of HNSCC, regarding perineural invasion, the number of nerves, the size of nerves or the different patterns of LHR-IM (48,49). In addition, by using these tools, the interobserver variability is reduced on morphological and immunohistochemical grounds (50), which is crucial with predictive biomarker evaluation, such as PD-L1 CPS. In this context, the performance of these models may exceed that of experienced pathologists.

One of the limitations of this study is the inclusion of both biopsy and surgical material, as well as the accurate representation of histologic characteristics in small biopsy samples. Biopsy material was more common in oropharyngeal tumors. In addition, histological parameters were mainly examined in whole sections, whereas immunohistochemical parameters were examined only on TMA sections. Further, site-specific subgroup analysis for single or profiled parameters in terms of OS was not possible due to the imbalanced respective group sizes. Male predominance may affect the conclusions of this study due to sex-related biological differences. Gender-specific histopathological parameters with potential clinical value weren't investigated and future studies examining sex as a potential modifier of prognostic outcomes in HNSCC are needed. Finally, the lack of HPV-positive tumors should be regarded as an additional limitation.

In conclusion, based on our findings, it seems that the malignant potential of HNSCC is reflected in several tumor biology-related parameters, such as CD8+ spatial heterogeneity, tumor center and/or invasive margin perineural invasion and LHR-IM. Furthermore, perineural invasion and LHR-IM are confirmed as independent histological risk factors in HNSCC. Integrating the above parameters in pathology reports may provide more accurate prognostic information for patients with HNSCC than the routinely assessed histologic grade. The herein presented site-specific and tumor compartment histological parameters may be included in larger studies with deep learning approaches, in order to enable revising histologic grade in HNSCC in a clinically meaningful way, as required in the context of personalized diagnostics.

Supplementary Material

Supporting Data

Supporting Data

Acknowledgements

The authors would like to thank Mrs. Maria Moschoni (data coordinator of HeCOG) for administrative support.

Funding

The study is part of the NCR-17-12885 project funded by Astra Zeneca and conducted by HeCOG. It was also partially supported by a Hellenic Society of Medical Oncology (HeSMO) grant (grant no. HE_TR5/25). The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript.

Availability of data and materials

The data generated in the present study may be requested from the corresponding author.

Authors' contributions

SET, VK, GK, PH, GF and TK conceptualized and designed the study, performed the formal analysis and wrote the original draft; KM, SP, KV, SC, AP, PH and GF collected and reviewed clinical and histopathological data. GF and SC confirm the authenticity of all the raw data. All authors have read and approved the final manuscript.

Ethics approval and consent to participate

Written informed consent was obtained from all patients for the use of their data and biological material for research purposes. The protocol was approved by the Bioethics Committee of the Aristotle University of Thessaloniki, School of Health Sciences, Faculty of Medicine (approval no. 5/18.12.2019).

Patient consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Glossary

Abbreviations

Abbreviations:

References