口腔鱗狀細胞癌免疫微環境的空間特徵、多重免疫組織化學與多光譜影像分析

Abstract

口腔鱗狀細胞癌 (Oral squamous cell carcinoma, OSCC) 為頭頸部鱗狀細胞癌中主要的亞型。在臺灣，嚼食檳榔是 OSCC 的重要危險因子，使得此疾病於臨床上造成沉重的負擔。然而，晚期OSCC的治療成效目前仍不理想。 傳統上使用的生物標記在預後及治療反應的預測上應用價值有限，其關鍵原因在於這些指標忽略了腫瘤微環境 (Tumor microenvironment, TME) 中免疫細胞的空間分布與結構限制，而這些空間特徵在調控對抗腫瘤免疫反應中扮演重要角色。基於此一限制，本研究建立了一套以組織為基礎、具備空間解析度的量化分析架構，用以評估免疫檢查點特徵 (Immune checkpoint signature, ICS)。透過此方法，我們進一步闡明 T 細胞上 ICS 的表現如何影響其功能狀態，並說明其與 OSCC 免疫治療反應率及臨床預後之間的關聯。 分析公開的單細胞轉錄體定序資料後，我們觀察到在接受免疫治療後，除惡性腫瘤細胞外，CD8⁺ T 細胞與多個免疫檢查點分子 (PD-1、TIGIT、LAG-3與TIM-3) 的表現密度上亦呈現顯著變化。因此，本研究採用多重免疫組織化學染色 (Multiplex immunohistochemistry, mIHC) 結合AI多光譜影像分析，以整合性評估CD8⁺ T 細胞的免疫檢查點特徵。研究主要聚焦於腫瘤免疫微環境中與類耗竭 (exhaustion-like) 狀態相關的抑制性路徑，並以 PanCK 定義腫瘤區域，同時量化腫瘤區域內 CD8⁺ T 細胞上 PD-1、TIGIT、LAG-3 和 TIM-3 的表現。每個樣本擷取 35 至 50 個區域 (Region of interests, ROIs) 進行分析，並整合 inForm 組織分割、單細胞表型分析與 SIMPiE 特徵萃取技術。最終，依據 CD8⁺ T 細胞的檢查點共表現模式，將其分類為不同的檢查點定義表型 (checkpoint-defined phenotypes)，並計算其在腫瘤區與間質區中的細胞密度 (cells/mm²)。 為評估免疫細胞的空間組織特徵，本研究採用最近鄰距離分析 (Nearest neighbor distance analysis) ，以量化 CD8⁺ 亞群與腫瘤區域之間的鄰近程度，並特別聚焦於 50 µm 的「活性區域」(Active zone)，此區域被認為是細胞之間最可能發生直接交互作用的空間範圍。基線圖譜 (Baseline profiling) 結果顯示，OSCC 的免疫群集呈現出顯著的空間分化特徵，並具有明確且異質的分布模式：部分由免疫檢查點定義的亞群主要侷限於間質區 (Stroma)，而另一些亞群則在腫瘤 (Tumor nests) 中呈現較高的盛行率。此種空間異質性顯示，單純依靠細胞總密度作為評估指標，並不足以完整反映腫瘤免疫脈絡。由於不同免疫亞群在腫瘤分艙之間呈現不均勻分布，其與惡性細胞發生直接交互作用的機會可能受到空間限制。 接下來，本研究建立一套以免疫治療反應後存活時間為基準的篩選策略 (Response-anchored screening strategy)，以處理 32 種初始檢查點定義表型所帶來的高度複雜性。在包含 18 位 OSCC 患者治療前後配對樣本的探索性隊列中，我們分析免疫亞群密度的治療前後變化量（Δ = Post − Pre）與臨床結果（性別、年齡、整體存活期及 TNM 分期）之間的相關性。此方法使我們得以優先鑑定兩種特定的細胞狀態（命名為ICS #1 與 ICS #2），其在反應者 (Responders) 與無反應者 (Non-responders) 之間呈現最顯著的統計差異。進一步的空間解析結果顯示，在反應者中，這些細胞狀態於接受抗 PD-1 治療後，其密度顯著增加，且在 50 µm 活性區域內呈現較高的空間富集程度；相較之下，無反應者則維持了免疫亞群與腫瘤之間的空間分離模式。這些結果顯示，特定免疫程式與腫瘤邊界之間的空間鄰近性與治療反應相關，並提供了獨立於免疫細胞數量之外的重要資訊。 最後，我們將上述空間特徵整合至一套臨床風險分數模型 (Risk score model，由 實驗室同仁Ho Nguyen Van Tan 開發)。該模型鑑定出一項由三個 ICS 衍生空間特徵構成的預後指標，能利用風險分數，精準對OSCC病人進行整體存活期及治療反應之評估，且該風險分數在排除吸菸史或檳榔暴露等潛在干擾因子後，仍維持穩定的預測效能。綜上所述，本研究建立了一套標準化的空間圖譜分析架構，證實結合免疫表型分析與腫瘤鄰近指標，可為 OSCC 免疫治療的個體化醫療提供具生物學基礎且具臨床可行性的分析工具。

ABSTRACT Oral squamous cell carcinoma (OSCC), a major subtype of head and neck squamous cell carcinoma, remains a substantial clinical burden in Taiwan where betel quid chewing is a dominant risk factor and treatment outcomes for advanced disease are still poor. Conventional biomarkers often provide limited predictive value, largely because they neglect the spatial topographies and structural constraints within the tumor microenvironment (TME) that fundamentally orchestrate antitumor immunity. The present study addresses this gap by developing a spatially resolved, tissue-based framework to quantify and evaluate the immune checkpoint signature (ICS). We demonstrate how ICS expression on T cells modulates their functional state, thereby influencing immunotherapy response rates and clinical outcomes in OSCC. Motivated by analyses of two publicly available single-cell RNA-sequencing datasets, we observed that, in addition to malignant tumor cells, CD8⁺ T cells exhibited significant density changes of four immune checkpoint molecules PD-1, TIGIT, LAG-3, and TIM-3 following immunotherapy treatment. Based on these findings, we employed multiplex immunohistochemistry (mIHC) combined with AI-assisted multispectral imaging to evaluate a focused immune checkpoint panel on CD8⁺ T cells. Specifically, we quantified the expression of PD-1, TIGIT, LAG-3, and TIM-3 on CD8⁺ T cells within PanCK-defined tumor compartments, targeting key inhibitory axes associated with exhaustion-like states in the TME. Between 35 and 50 regions of interest (ROIs) per sample were acquired via multispectral imaging and processed through a standardized pipeline utilizing inForm-based tissue segmentation and single-cell phenotyping, followed by SIMPiE-based feature extraction. CD8⁺ T cells were stratified into checkpoint defined phenotypes based on co-expression patterns, quantified as compartment-resolved densities (cells/mm²) in tumor and stroma. To evaluate the spatial organization of these immune cells, we employed nearest-neighbor distance analysis to quantify the proximity of CD8⁺ subsets to tumor regions, specifically focusing on a 50 µm active zone where cellular interactions are most likely to occur. Our baseline profiling demonstrated that the OSCC immune landscape is significantly compartmentalized, characterized by distinct distribution patterns where certain checkpoint defined subsets are predominantly restricted to the stroma while others exhibit higher prevalence within tumor nests. This spatial heterogeneity underscores that total cell density alone cannot fully capture the immune context, as the disproportionate distribution of specific subsets between compartments may limit their potential for direct interaction with malignant cells. A central contribution of this study is the response-anchored screening strategy implemented to navigate the complexity of the 32 initial checkpoint defined phenotypes. Utilizing a discovery cohort of 18 OSCC patients with paired Pre- and Post-immunotherapy (IO) samples, we performed correlation analyses between clinical outcomes including sex, age, overall survival, and TNM stage and the dynamic shifts in immune subsets, calculated as the delta (Post-Pre) change in density. This approach allowed us to prioritize two specific cell states (ICS #1 and ICS #2) that exhibited the most significant statistical divergence between responders and non-responders. Further spatial interrogation revealed that in responding patients, these nominated states significantly increased in density and became more concentrated within the 50 µm active zone following anti-PD-1 therapy. Conversely, non-responders maintained a pattern of spatial separation between these immune subsets and the tumor. These findings demonstrate that the physical proximity of specific immune programs to the tumor margin is a critical dimension of therapeutic response that provides information independent of simple cell abundance. Finally, we integrated these spatial features into a clinical risk score model (developed by Ho Nguyen Van Tan). This framework identified a compact signature of three ICS-derived features that accurately stratified overall survival and treatment response in both discovery and independent validation cohorts. Notably, this risk score maintained its prognostic value regardless of smoking history or betel nut exposure. Collectively, this study establishes a standardized spatial profiling framework, demonstrating that combining immune phenotyping with tumor-proximity metrics provides a biologically grounded and clinically actionable tool for personalizing immunotherapy in OSCC.