The Prognostic Role of CD8⁺ Tumor‐Infiltrating Lymphocyte Density in Head and Neck Mucosal Melanoma

Abstract

Objective

To explore the prognostic role of CD8+ tumor‐infiltrating lymphocytes (TIL) density in head and neck mucosal melanoma (HNMM) using an automated, quantitative, immunohistochemistry‐based methodology.

Study Design

Retrospective cohort study.

Setting

Single‐institution, tertiary care hospital.

Methods

Forty‐five patients with primary HNMM treated between January 2002 and December 2021 at the University of Pittsburgh were included. For patients with ample primary tumor specimens, CD8⁺ TIL density was quantified using an automated image analysis algorithm on digitized slides. Receiver operating characteristic curves were used to delineate a biomarker cutoff. Survival analysis using the Kaplan‐Meier method and Cox proportional hazards was performed with recurrence‐free survival (RFS) and overall survival (OS) as endpoints.

Results

Thirty‐four patients had adequate pathologic specimens available for analysis. CD8⁺ TIL density threshold of 373.6 cells/mm² best delineated patients based on survival outcomes. On log‐rank testing, TIL high patients trended toward improved 5‐year RFS (p = 0.06) and OS (p = 0.074) when compared to TIL low patients. Using multivariate analysis, high TIL density was significantly associated with improved RFS (hazard ratio [HR] (95% confidence interval [CI]): 0.17 (0.03‐0.83), p = 0.028) but not improved OS (HR (95% CI): 0.48 (0.08‐2.90), p = 0.4). Tumor characteristics such as advanced nodal staging (HR (95% CI): 13.2 (1.68‐104), p = 0.014), and presence of metastatic disease (HR (95% CI): 48.2 (1.94‐1196), p = 0.018) were associated with worsened RFS.

Conclusion

Automated, digital CD8⁺ TIL density enumeration is a reliable method of assessing TIL density in HNMM patients. High CD8⁺ TIL density is an independent factor associated with improved RFS and trended toward improved OS.

Mucosal melanomas (MM) are a rare subtype of melanoma which arise from mucosal epithelium—most commonly in the head and neck. ¹ It has emerged as a distinct clinical entity from cutaneous melanoma with worse prognosis and more aggressive disease behavior. ² , ³ , ⁴ , ⁵ , ⁶ The advent of immunotherapies such as immune checkpoint inhibitors (ICIs) has revolutionized the treatment landscape of melanoma. However, response rates in MM are significantly worse (~19‐23%) when compared to cutaneous melanoma, suggesting that there are inherent differences in disease biology affording therapeutic resistance. ⁷ , ⁸ , ⁹

Tumor‐infiltrating lymphocytes (TILs) are a well‐recognized component of the host immune response to tumorigenesis. The density and distribution of TILs are thought to reflect the underlying immunogenicity of the tumor microenvironment. Within tumors, tumor‐antigen‐specific CD8⁺ T‐cells have a diminished ability to proliferate, to produce cytokines, and to induce tumor cell lysis. ¹⁰ , ¹¹ , ¹² Immunotherapies such as immune checkpoint inhibitors (ICI) act to enhance the proliferation and lytic functions of subdued intra‐tumoral, tumor‐antigen‐specific CD8⁺ T‐cells. ¹³ , ¹⁴ , ¹⁵ As such, numerous studies have demonstrated that robust lymphocytic infiltration is an independent prognostic marker for improved survival in a variety of solid tumor types, such as ovarian, colorectal, and cutaneous melanoma. ¹⁶ , ¹⁷ , ¹⁸ Because of the rarity of disease, there has been limited research on the prognostic role of CD8⁺ TIL density as a prognostic marker in head and neck mucosal melanoma (HNMM).

Due to poor response rates to multi‐modal therapies, it has become increasingly important to develop easily reproducible and accurate prognostic models of mucosal melanoma. To date, there is no standardized method to assess CD8⁺ TIL density. Previous work in metastatic cutaneous melanoma demonstrated that an automated algorithm for CD8⁺ TIL density enumeration was a reliable and effective method. ¹⁸ Automated image analysis for CD8⁺ TIL density quantification offers reproducible, observer‐independent assessment that is readily scalable to high‐throughput workflows, overcoming a key limitation of manual enumeration methods. For this reason, the present study aims to add to the growing literature by evaluating the relationship between CD8⁺ TIL density and clinical outcomes in a single‐institution cohort of patients with HNMM using a previously validated, automated image analysis algorithm. Using the same method demonstrated in cutaneous melanoma, we hypothesize that higher TIL density is independently associated with improved overall and recurrence‐free survival.

Methods

Study Design

This is a retrospective study including patients from a single tertiary care institution with primary HNMM who were treated with primary surgical resection or excisional biopsy prior to initiation of definitive therapy between January 2005 and December 2022. Forty‐five patients met the diagnostic criteria. Of these, thirty‐four patients had sufficient volume of banked tumor specimen for immunohistochemistry and TIL density enumeration. Because this was a retrospective cohort study of a rare disease, an a priori sample size calculation was not feasible. All consecutive patients meeting inclusion criteria were included in order to maximize statistical power and minimize selection bias. Recurrence‐free survival (RFS) was defined as the time from the date of confirmed tissue diagnosis of MM to the date of radiographic or clinical evidence of recurrence. Overall survival (OS) was defined as the time from the date of confirmed tissue diagnosis of MM to the date of death. This study was approved by the University of Pittsburgh Institutional Review Board.

Histopathology and Digital Quantitative Image Analysis

Four μm tissue sections from representative tumor blocks were sectioned and mounted on positively charged glass slides. Tissue blocks were chosen from tissue obtained during primary surgical resection or from excision biopsies performed prior to initiation of any systemic therapies. CD8 immunohistochemistry (IHC) was performed using C8/144B clone antibodies at a concentration of 1:40 (catalog number: GA62361‐2; Agilent Dako). DAB (3,3ʹ‐diaminobenzidine) was then used as the color reagent for visualization. Negative reagent controls were performed using Monoclonal rabbit IgG antibody (Catalogue number: 760‐1029, Roche Diagnostics) on replicate samples.

Following IHC staining for CD8, the whole slide images were digitized using an Aperio AT 2 scanner set at a magnification of 40× and uploaded to eSlideManager (eSM). Digitized tissue sections were then annotated by designated areas of the slide occupied by tumor. This process was previously described for CD8 analysis in oropharyngeal squamous cell carcinoma and cutaneous melanoma. ¹⁸ , ¹⁹ A previously validated imaging algorithm was then used (Aperio Technologies, Inc.) to identify and count CD8⁺ cells within the annotated tumor area. TIL density was then calculated by dividing the number of CD8⁺ cells by the total area of annotated tissue analyzed. TIL densities across several (at least 3) separate representative cross‐sectional slides of the tumor were averaged for final reporting. Examples of the representative slide images and algorithmic enumeration of CD8⁺ TILs generated by this process can be found in a previously published manuscript. ¹⁹

Statistical Analysis

All statistical analyses were performed using RStudio (Version: 2025.09.0 + 387; Posit Software). Demographic and clinicopathologic features of the cohort population were reported using descriptive statistics. Continuous variables were represented as a mean with standard deviation (SD). Comparisons of categorical variables were performed using Fischer Exact testing, when appropriate. Threshold values for CD8⁺ TIL density were determined based on the point of maximal area under the curve on a plotted receiver operating characteristic (ROC) curve. This threshold value was then used to separate the cohort into high and low CD8⁺ TIL density samples. The primary endpoint of this study was 5‐year OS, and the secondary endpoint was 5‐year RFS. The relationship between predictor variables and survival endpoints were then evaluated using Cox proportional hazard models. Kaplan‐Meier survival curves and logrank testing were then used to visualize OS and RFS by high and low TIL density.

Results

Forty‐five patients met the diagnostic criteria of primary HNMM for inclusion in this study. The mean age at diagnosis was 72.5 years (SD: 11.6 years). Most patients were male (62.2%), and a vast majority were Caucasian (91.1%). The mean follow‐up time was 39 months. The most common primary site of the tumor was the nasal cavity/paranasal sinuses (82.2%) followed by the oral cavity (13.3%). By AJCC 8^th edition TNM staging, most patients presented with Stage III (47.7%) or Stage IVa (34.1%) disease. Demographics and histopathologic characteristics of the cohort are summarized in Tables 1 and 2, respectively. All patients underwent primary resection with curative intent, with 44.4% of patients receiving radiation therapy, 51.1% receiving immunotherapy, and 4.4% receiving chemotherapy. Immunotherapies included immunomodulator therapies such as interferon‐alpha or ICIs such as CTLA4 and PD‐(L)1 inhibitors.

Table 1.

Demographic and clinical characteristics

Variables (n = 45)	n (%)
Gender
Female	17 (37.8)
Male	28 (62.2)
Race
Caucasian	41 (91.1)
African American	1 (2.2)
Asian	3 (6.6)
Age at diagnosis (years)
Mean (SD)	72.5 (11.6)
Follow up time (months)
Mean (SD)	39 (40.3)
Smoking history (n = 40)
Never	23 (57.5)
Current	7 (17.5)
Former	10 (25)
Alcohol history (n = 39)
No	22 (56.4)
Yes	17 (43.6)
Radiation therapy
No	25 (55.6)
Yes	20 (44.4)
Chemotherapy
No	43 (95.6)
Yes	2 (4.4)
Immunotherapy
No	22 (48.9)
Yes	23 (51.1)

Abbreviations: n, sample size; SD, standard deviation.

Table 2.

Histopathologic characteristics

Variables (n = 45)	n (%)
T stage
T3	24 (53.3)
T4a	17 (37.8)
T4b	4 (8.9)
N stage (n = 44)
N0	32 (72.7)
N1	12 (27.3)
M stage (n = 44)
M0	40 (90.9)
M1	4 (9.1)
AJCC 8th ed. TNM staging (n = 44)
III	21 (47.7)
IVa	15 (34.1)
IVb	3 (6.8)
IVc	5 (11.4)
Primary site
Nasal cavity	37 (82.2)
Nasopharynx	1 (2.2)
Oral cavity	6 (13.3)
Oropharynx	1 (2.2)
Margin status (n = 25)
Negative	18 (72)
Positive	7 (28)
Lymphovascular invasion (n = 22)
No	14 (63.6)
Yes	8 (36.4)
Perineural invasion (n = 21)
No	14 (66.7)
Yes	7 (33.3)
TIL density (n = 34)
Mean (SD)	334.7 (253.5)

Abbreviations: n, sample size; SD, standard deviation.

Thirty‐four patients had sufficient volume of tumor specimen available for immunohistochemical analysis. The mean calculated TIL density for these patients was 334.7 cells/mm². A ROC curve was then drawn to identify a CD8⁺ TIL density cutoff value which would reliably segregate survivors and non‐survivors (Figure 1). Using the ROC curve, a threshold value of 373.6 cells/mm² was chosen with an area under the curve (AUC) of 0.725, indicating an acceptable discriminative performance. Twelve patients with TIL densities above the threshold were labeled as high TIL density. Whereas the remaining 22 patients with TIL densities below the threshold were designated as low TIL density. At 5 years, 7 of the 12 patients with high TIL densities were alive. In contrast, only 3 of the 22 patients with low TIL densities were alive at the 5‐year mark. There was no statistically significant difference in rates of adjuvant therapies between TIL high and TIL low patients (radiation [TIL High vs TIL Low]: 53.8% vs 36.4%, p = 0.29; chemotherapy: 8.3% vs 4.5%, p = 1.0; immunotherapy: 33.3% vs 45.5%, p = 0.72).

Figure 1.

Receiver operating characteristics (ROC) curve generated based on CD8⁺ tumor‐infiltrating lymphocytes density. Threshold of 373.6 cells/mm² was chosen with the highest area under the curve of 0.725.

On univariate analysis, high TIL density was not significantly associated with improved OS when compared to low TIL density, but trended toward significance (hazard ratio [HR]: 0.43, 95% confidence interval [CI]: 0.17‐1.11, p = 0.08). Similarly, trends were observed for high TIL density and RFS (HR: 0.5, 95% CI: 0.23‐1.04, p = 0.065). The results of our univariate analyses for OS and RFS are summarized in Table 3. On multivariate analysis, high TIL density was independently associated with improved RFS (HR (95% CI): 0.17, 95% CI: 0.03‐0.83, p = 0.028) after controlling for T stage, N stage, M stage, and perineural invasion. Tumor characteristics such as advanced nodal staging (HR (95% CI): 13.2 (1.68‐104), p = 0.014), and presence of metastatic disease (HR (95% CI): 48.2 (1.94‐1196), p = 0.018) were associated with worsened RFS (Table 4). High TIL density patients trended toward an improved 5‐year RFS (p = 0.06) on Kaplan‐Meier curves when compared to low TIL density patients (Figure 2).

Table 3.

Univariate Cox proportional hazards model

Predictor	Overall survival		Recurrence‐free survival
	HR (95% CI)	p‐value	HR (95% CI)	p‐value
Site of primary tumor
Nasal cavity	‐		‐
Nasopharynx	0.96 (0.13‐7.2)	>0.9	0.72 (0.10‐5.32)	0.7
Oral cavity	1.65 (0.56‐4.9)	0.4	1.69 (0.63‐4.51)	0.3
Oropharynx	2.17 (0.29‐16.5)	0.5	0.96 (0.13‐7.11)	>0.9
T stage
T3	‐
T4a	1.98 (0.93‐4.21)	0.077	1.85 (0.94‐3.62)	0.074
T4b	2.07 (0.59‐7.26)	0.3	1.88 (0.62‐5.68)	0.7
N stage
N0	‐
N1	2.17 (0.97‐4.84)	0.058	1.99 (0.98‐4.04)	0.057
M stage
M0	‐
M1	2.19 (0.65‐7.42)	0.2	3.52 (1.11‐11.1)	0.032
Perineural invasion
No	‐		‐
Yes	7.23 (1.90‐27.5)	0.004	3.10 (0.98‐9.76)	0.053
Tumor TIL density
Low	‐		‐
High	0.43 (0.17‐1.11)	0.081	0.5 (0.23‐1.04)	0.065
Lymphovascular invasion
No	‐		‐
Yes	1.02 (0.32‐3.30)	>0.9	0.86 (0.29‐2.52)	0.8
Margin status
Negative	‐		‐
Positive	1.6 (0.55‐4.63)	0.4	1.56 (0.58‐4.18)	0.4
Immunotherapy treatment
No	‐		‐
Yes	0.41 (0.19‐0.88)	0.022	0.71 (0.38‐1.31)	0.3

Abbreviations: CI, confidence intervals; HR, hazard ratio; TIL, tumor‐infiltrating lymphocytes.

Table 4.

Multivariate Cox proportional hazards model for recurrence‐free survival

Predictor	Recurrence‐free survival
	HR (95% CI)	p‐value
T stage
T3
T4a	0.43 (0.04‐4.21)	0.5
T4b	0.57 (0.02‐16.8)	0.7
N stage
N0
N1	13.2 (1.68‐104)	0.014
M stage
M0
M1	48.2 (1.94‐1196)	0.018
Perineural invasion
No	‐
Yes	2.62 (0.18‐38.1)	0.5
Tumor TIL density
Low	‐
High	0.17 (0.03‐0.83)	0.028

Note: Univariate p value < 0.1 was used as inclusion criteria.

Abbreviations: CI, confidence intervals; HR, hazard ratio; TIL, tumor‐infiltrating lymphocytes.

Figure 2.

High versus low CD8⁺ tumor infiltrating lymphocytes (TIL) density in recurrence‐free survival in patients with head and neck mucosal melanoma. Log‐rank testing between the two survival curves generated a p‐value of 0.06.

For OS, on multivariate analysis, high TIL density (HR: 0.48, 95% CI: 0.08‐2.90, p = 0.4) was not associated with improved 5‐year OS when compared to low TIL density patients after controlling for T stage, N stage, M stage, perineural invasion, and immunotherapy treatment status (Table 5). On log‐rank testing, high TIL density patients trended toward improved 5‐ year OS (p = 0.074) when compared to low TIL density patients (Figure 3).

Table 5.

Multivariate Cox proportional hazards model for overall survival

Predictor	HR (95% CI)	p‐value
T stage
T3	‐
T4a	1.29 (0.23‐7.08)	0.8
T4b	6.85 (0.16‐299)	0.3
N stage
N0	‐
N1	5.14 (0.41‐64.6)	0.2
Perineural invasion
No	‐
Yes	1.40 (0.07‐26.3)	0.8
Tumor TIL density
Low	‐
High	0.48 (0.08‐2.90)	0.4
Immunotherapy treatment
No	‐
Yes	1.13 (0.25‐5.21)	0.9

Note: Univariate p value < 0.1 was used as inclusion criteria.

Abbreviations: CI, confidence intervals; HR, hazard ratio; TIL, tumor‐infiltrating lymphocytes.

Figure 3.

High versus low CD8⁺ tumor infiltrating lymphocytes (TIL) density in overall survival in patients with head and neck mucosal melanoma. Log‐rank testing between the two survival curves generated a p‐value of 0.074.

Discussion

In this retrospective cohort study of patients with head and neck mucosal melanoma, we demonstrated a low‐cost and reliable process to evaluate TIL density in primary tumor resection specimens. This workflow was previously validated in metastatic cutaneous melanoma. ¹⁸ Using this method, we were able to derive a threshold value for TIL density of 373.6 cells/mm², which best segregated survivors and non‐survivors. We found that a high TIL density was significantly associated with improved RFS and showed similar promising trends in improved OS when compared to low TIL density patients.

There is a strong biological basis for using TIL density as a prognostic marker. Within the tumor microenvironment, immune cells such as macrophages, dendritic cells, CD4⁺ T helper cells, CD8⁺ cytotoxic T cells and natural killer T cells each play an integral role in the host mounted response to tumorigenesis. ²⁰ CD8⁺ cytotoxic T cells are fundamental to the recognition and elimination of tumor cells and are well recognized primary drivers of therapeutic potency. A tumor's ability to induce CD8⁺ T cell tolerance is central to their ability to subvert the normal immune response. ²¹ Consequently, the balance between T cell activation and suppression within the tumor microenvironment is thought to be a determining factor in clinical response to therapy. ²⁰ Thus, a dense intra‐tumoral lymphocytic infiltrate reflects a more robust antitumoral response that can be leveraged for therapeutic purposes. ¹⁷ , ²² , ²³ This is best demonstrated with the use of ICI therapies. The receptor/ligand targets of ICI therapies—CTLA‐4 and PD‐(L)1—typically function to inhibit T cell activation and CD8⁺ T‐cell‐mediated tumor lysis. Through inhibition of these pathways, ICIs promote the lytic function of subdued intra‐tumoral, tumor‐antigen‐specific CD8⁺ T cells. ¹⁰ , ¹¹ , ¹²

Our study adds to the growing body of literature examining the prognostic value of TIL density in HNMM. Ledderose et al. and Guo et al. were two of the earliest studies examining TIL density in SNMM using a previously validated, three‐tier subjective assessment—absent, non‐brisk, and brisk—as a prognostic biomarker for survival. ²⁴ , ²⁵ , ²⁶ Guo et al. found that brisk TIL density was significantly associated with improved 3‐year recurrence‐/metastasis‐free survival (p = 0.007) and trends toward improved 3‐year OS (p = 0.07) when compared to non‐brisk TIL patients. ²⁵ Ledderose et al. reported similar findings showing significant improvements in both RFS and OS in patients with brisk TIL infiltrated when compared to absent TIL infiltration. ²⁶ In a follow‐up study, Ledderose et al. demonstrated, using a new semi‐quantitative method, that high CD3⁺ and CD8⁺ TIL populations were independent predictors of improved OS but not RFS. ²⁷ Akin to the findings of the other studies, our work demonstrated that high CD8⁺ TIL density was independently associated with an 87% reduction in the risk of 5‐year recurrence (HR: 0.17, 95% CI: 0.03‐0.83, p = 0.028) and trended toward improved OS survival (p = 0.074) when compared to low TIL density patients. In aggregate, these studies highlight the potential of TIL density as a promising prognostic factor in HNMM.

Furthermore, our study presents the first time a validated, fully quantitative assessment model of TIL density has been used in HNMM. The use of IHC allows for this methodology to be more targeted and precise in its assessment of TIL density than the previously used qualitative and semi‐quantitative models. Moreover, this methodology could also allow us to quantify a multitude of different cell populations within the tumor microenvironment in one streamlined workflow. There is growing evidence in other solid tumors which suggest that both the spatial organization of immune infiltration and proximity to other cell populations have prognostic value and warrant continued investigation. ²⁸ , ²⁹ , ³⁰ , ³¹ , ³² Early studies have observed that areas of immune‐exclusion within tumors can be predictive of poor response to therapy and that their creation may be driven by cancer‐associated fibroblasts which act to both hinder T cell infiltration from the periphery through stromal barrier formation and recruit immunosuppressive cells through chemokine‐mediated enrichment. ³³ , ³⁴ The ability to analyze the spatial relationships of immune cell populations between one another, to tumor, and/or non‐neoplastic stroma using a cost‐effective and reproducible technique will become invaluable as we continue to refine our prognostic models for HNMM. Early studies examining correlations between TIL density and molecular expression of immune checkpoint molecules has generated mixed results. ³⁵ , ³⁶ Nevertheless, correlating TIL density with tumor mutational burden or expression of immune checkpoint receptors/ligands (e.g. PD‐(L)1, TIM‐3, TIGIT, etc.) could help identify biologically distinct subgroups with differential therapeutic susceptibilities. ³⁷

Limitations of our study include its design as a retrospective case‐control study from a single‐institution with post hoc assessment of CD8⁺ TIL density. Consequently, the design of our study limits the generalizability of our findings to other treatment settings and institutions. Because CD8⁺ TIL density was not calculated at the time of treatment and was not used to guide management, our study also has a limited role in determining how TIL density should be best utilized to guide future management of HNMM. Due to the rarity of disease, our small sample size also limited our statistical power. Lastly, by requiring sufficient tumor specimens for immunohistochemical analysis, we may have inadvertently excluded patients whose tumors were very small in volume or too large to be surgically resectable. Despite these limitations, our findings highlight the importance of understanding the tumor microenvironment in HNMM.

In conclusion, this study is the first to evaluate the prognostic value of CD8⁺ TIL density in HNMM using an automated, quantitative method. Higher CD8⁺ TIL density is independently associated with improved 5‐year RFS in HNMM, with trends to suggest similar findings in 5‐year OS. Due to the rarity of disease, multi‐institutional, prospective studies are warranted to validate our results. These future studies will help guide standardized approaches to integrating CD8⁺ TIL density into the treatment decision making for HNMM.

Author Contributions

Nicholas K. Fung, conceptualization/design, methodology, data procurement, data analysis, manuscript writing, funding procurement, presentation; Douglas Hartman, conceptualization/design, methodology, data procurement, data analysis, manuscript review; Diwakar Davar, conceptualization/design, methodology, data analysis, manuscript review; Eric W. Wang, conceptualization/design, methodology, data analysis, manuscript writing, funding procurement.

Disclosures

Competing interests

Nicholas Fung: Consultant for Allergy Education Partners, LLC. Douglas Hartman: None. Diwakar Davar: Grants from GSK, Arcus Biosciences, Immunocore, Merck, and Regeneron Pharmaceuticals Inc.; personal fees from ACM Biolabs, ABL Bio, Ascendis Pharma, Clinical Care Options, Gerson Lehrman Group, Med Learning Group, Merck, Replimune, TriSalus Life Sciences, Xilio Therapeutics, and Castle Biosciences; and other support from ACM Biolabs, Immunocore, Replimune, mbiomics, and Zola Therapeutics outside the submitted work. Eric Wang: Consultant for Stryker.

Funding source

This work was supported by an AAO‐HNSF Resident CORE Research Grant awarded to N. K. F. in 2022.