Novel Pathological Predictive Factors for Extranodal Extension in Oral Squamous Cell Carcinoma: a Retrospective Cohort Study Based on Tumor Budding, Desmoplastic Reaction, and Tumor-inltrating Lymphocytes

Extranodal extension (ENE) is a poor prognostic factor for oral squamous cell carcinoma (OSCC). Identifying ENE by clinical and/or radiological examination is dicult, thereby leading to unnecessary neck dissections. Currently, no denitive predictors are available for ENE. Thus, we aimed to determine the histological predictors of ENE by routine histopathological examination using biopsy and surgically resected specimens. Methods: This retrospective study included 186 surgically resected OSCC and 83 matched biopsy specimens. Clinical features associated with the tumor microenvironment, including desmoplastic reaction (DR), tumor budding (TB), and tumor-inltrating lymphocytes (TILs), were evaluated using hematoxylin and eosin-stained primary OSCC and neck dissection specimens. These histological features were divided into two groups: DR-immature (DR-I) and DR-mature (DR-M); TB-high (TB-H) and TB-low (TB-L); and TILs-low (TILs-L) and TILs-high (TILs-H). Clinical depth of invasion (cDOI) and pathological DOI (pDOI) were adapted for biopsies and resections, respectively; DOI was evaluated as DOI >10 mm and DOI ≤ 10 mm. The clinicopathological relationships between these histopathological features and ENE and the independent risk factors for ENE were analyzed. The histological predictors of ENE were evaluated. Results: The histological status of DR, TILs, and TB present in biopsy and resection specimens showed high accuracy with that of ENE. DR-I, TILs-L, and TB-H were signicantly associated with lymph node metastasis, cDOI, and pDOI. Bivariate and multivariate analyses revealed that TB-H and pDOI >10 mm in resections were independent factors for the presence of ENE (ENE+). The combination of TB-H/pDOI >10 mm in resection specimens showed high specicity (91%) and accuracy (83%) regarding ENE+. Although there proved to be no independent factors in biopsies, DR-I and TILs-L were signicantly associated with ENE+ (p<0.001). The combination of DR-I/TILs-L/cDOI >10 mm in biopsies exhibited high sensitivity and specicity with ENE+ (70% and 77%, respectively, p<0.001). These histological predictors could detect even minor ENE (<2 mm). >10 resections biopsies. TILs are a selected population of T-cells with a higher specic immunological reactivity against cells biopsies, patients with OSCC. Further studies with a large number of patients are needed to establish the signicance of DR, TB, and TILs, especially in biopsies, in patients with OSCC and to establish a useful and versatile nomogram using histopathological factors in biopsies for evaluating the presence of ENE in patients with OSCC.

The status of the three pathological features in the ENE foci was also evaluated using the H&E-stained slides of the cervical neck dissection samples. All the H&E-stained slides used in this study were prepared for routine pathological examination. These histological factors were examined for OSCC in the biopsy specimens using a single slide, the resected specimens including the deepest part of the tumor, and the neck dissected specimens including the major ENE part in lymph node metastasis (Figures 1b, c). The score of the three pathological features was evaluated by two pathologists (NY and IM).
The DR was histologically classi ed as immature (DR-I) or mature (DR-M) [24]. DR-I refers to a brotic stroma with myxoid changes as observed across a microscopic eld under a ×40 objective (Figure 2a), and DR-M refers to the absence of a myxoid stroma or presence of keloid-like collagen lacking mature stroma (Figure 2b) [24]. The stromal TILs were evaluated as the average number of lymphocytes composed of stromata as observed under ×20 to ×40 objectives; they were categorized as low (TILs-L ≤20%) or high (20% <TILs-H) (Figures 2a-c) [19]. TB was assessed and classi ed as low (TB-L ≤10) or high (10 >TB-H) as observed under a ×20 objective (Figures 2d-f) [25]. The DOI of primary OSCC was de ned as the perpendicular distance between the extent of deep tumor invasion to the basement membrane of the adjacent mucosa [6]. The DOI was subdivided into the following groups: DOI ≤5 mm, DOI >5 mm, DOI ≤10 mm, and DOI >10 mm [6]. cDOI was measured by MRI ndings and pDOI was measured histologically for analyzing the clinicopathological features and ENE risk factors of the biopsy and resection specimens, respectively. pT1 and pT2 OSCC cases were classi ed as early OSCC cases and pT3 and pT4 as progressive OSCC cases.

Statistical analyses
Patient characteristics were compared between the two groups in terms of DR-I and DR-M, TB-H and TB-L, TILs-H and TILs-L, cDOI for biopsies (≤10 mm and >10 mm), and pDOI for resections (≤10 mm and >10 mm), using the chi-square or Fisher's exact test for categorical data. Bivariate logistic regression analysis was used to assess the relationship between the predictor variables and presence of ENE. A multivariate logistic regression model was constructed using the forward selection method. The diagnostic value of the risk factors was assessed by calculating sensitivity, speci city, positive predictive value (PPV), negative predictive value (NPV), and accuracy. The Mann−Whitney U test was used to analyze the differences in the diameters of the extent of ENE and tumor deposit between the combination risk factors. The analysis was performed using SPSS (version 20.0; IBM Corp., Armonk, NY, USA). Statistical signi cance was set at p≤0.05.

Results
Patient characteristics of the surgical specimens of OSCC A total of 674 patients were included in this study; 488 patients were excluded, and 186 who met the criteria were subjected to further analyses. The  Association of the clinicopathological features with DR, TB, and TILs in the resected specimens Table 2 shows the association of the clinicopathological features with DR, TB, and TILs in the resected specimens. DR-I, TB-H, and TILs-L were signi cantly associated with pDOI >10 mm (p<0.001, 0.01, and <0.001, respectively) and pT stage (1, 2 versus 3, 4) (p<0.001, 0.01, and <0.001, respectively). These three factors were also signi cantly associated with the presence of lymphovascular invasion and lymph node metastasis (all p<0.05). Moreover, in ltrative growth was signi cantly associated with DR-I, TB-H, and TILs-L (p<0.001, 0.01, and <0.01, respectively). However, tumor location was not signi cantly associated with DR, TB, or TILs. Association among DR, TB, and TILs in the resected specimens and ENE The results of the association among DR, TB, and TILs in the resected specimens and ENE sites are shown in Table 3. These three pathological features of the resection specimens were signi cantly associated with those of the ENE sites (all p<0.05).
Based on the clinicopathological analysis of DR, TB, and TILs in the resected specimens and association analysis with ENE, it was indicated that evaluating DR, TB, and TILs in the primary sites, as well as from ENE, could predict tumor behavior. Association of the clinicopathological features with DR, TB, and TILs in biopsy specimens Table 4 summarizes the association of clinicopathological features with DR, TB, and TILs in biopsy specimens. DR-I and TILs-L were signi cantly associated with cDOI >5 mm, cDOI >10 mm, and pT stage (all p<0.01) similar to that in resected specimens, although TB was not signi cantly associated. DR-I and TILs-L were also signi cantly associated with the presence and number of lymph node metastases (all p<0.05). TB-H was signi cantly associated with the number of metastatic lymph nodes (p=0.02). DR, TB, and TILs were not signi cantly associated with tumor location.  Association among DR, TB, and TILs in biopsies and ENE sites Similar to resections, the three pathological features of biopsies were signi cantly associated with those of the ENE site (all p<0.05, Table 5). In addition, these three features in biopsies were also signi cantly associated with matched resections (all p≤0.001, Table 6).

Discussion
In this study, we clearly demonstrated that the combination of TB-H/pDOI >10 mm in resections showed the highest speci city and accuracy for detecting ENE and that the combination of DR-I/TILs-L/cDOI >10 mm in biopsies showed high sensitivity, speci city, and accuracy for detecting ENE, including ENEmi. These results were derived from DR, TB, and TILs in biopsy specimens; resected specimens; and ENE site, all of which were signi cantly associated. Moreover, DR-I, TB-H, and TILs-L in resected specimens and DR-I and TILs-L in biopsy specimens were signi cantly associated with DOI >10 mm, pT stage, and lymph node metastasis, and TB-H and pDOI >10 mm in resected specimens were identi ed as independent factors for the presence of ENE in the multivariate analysis. We con rmed that evaluating the DR, TB, TILs, and DOI in the primary site and a combination of these factors in both resection and biopsies is useful for predicting the tumor behavior in metastatic lymph nodes and presence of ENE in patients with OSCC.
ENE occurs in 37.5−56.4% of lymph node metastases in OSCC [37,38]. Assessing ENE is vital during both pre-and postoperative examinations to decide whether wide-neck resection or chemoradiotherapy should be performed [39]. However, several ENEs are overlooked during clinical examinations due to limited radiological accuracy and the absence of a critical predictor of ENE. Recent studies have shown that TP53 mutation and SERPINE1 expression in broblasts increase the risk of ENE [40,41]. However, their abilities as predictors of ENE are limited, and the evaluation of these factors is di cult in routine work [40,41]; thus, a versatile and more simple method for supporting the cENE evaluation and predicting the presence of ENE using biopsy or surgically resected specimens of OSCC is required.
To the best of our knowledge, this study is the rst to demonstrate that TB-H, DR-I, and TILs-L and their combination in resections or biopsies are useful predictors for the presence of ENE, including ENEmi. DR, TB, and TILs at the ENE site were signi cantly correlated with those at the primary sites. This suggests that the TME in the primary sites is similar to that in the ENE sites; the TME affects the tumor behavior in OSCC and indicates the presence of ENE. In other words, histological evaluation of DR, TIL, and BD in the biopsy could predict the tumor behaviors in the resections and ENE.
The TME in the primary site might in uence the microenvironment in ENE [17,41]. During carcinogenesis, the synthesis and remodeling of TME components may induce cancer immuno-response and epithelial-mesenchymal transition, which promotes the miniaturization of tumor nests and deposition of reactive stroma [17,18]. Thus, the TME in the primary sites of OSCC might be related to the presence of lymph node metastasis and ENE [17,32]. Therefore, additional large-sample studies are needed to establish a nomogram (parameters including TB, DR, TILs, and cDOI in biopsies) to predict the presence of ENE in patients with OSCC, which can aid in providing appropriate treatment, including wide-neck lymph node dissection and/or chemoradiation therapy.
To the best of our knowledge, no reports have yet addressed the association between TB and ENE. This study clearly showed that TB-H was signi cantly associated with lymph node metastasis and was a powerful independent factor for ENE, showing high speci city in resections but not in biopsies. It is presumed that biopsies do not contain adequate invasion front regions, leading to low prediction accuracy of ENE. TB is de ned as a single cancer cell or a small cluster comprising less than ve cancer cells at the tumor invasion front [21][22][23]. Thus far, several studies have demonstrated that TB is a valuable prognostic marker, especially in the colon, nasopharynx, esophagus, lung, and breast [21, 22, 25-30, 32, 33]. A meta-analysis revealed that high-grade TB is signi cantly associated with lymph node metastasis in OSCC [21]. These are consistent with our results.
In contrast to TB, DOI in both biopsies and resections showed high sensitivity and low speci city for the presence of ENE, and this study is the rst to report that pDOI >10 mm is an independent risk factor for ENE. DOI is de ned as the extent of the tumor below the epithelial membrane, and it can be measured by preoperative radiological assessment using MRI or ultrasonography [6]. High tumor DOI is associated with a high frequency of lymph node metastasis and especially DOI >10.5 mm is associated with a high incidence of ENE [35]; moreover, the high correlation between cDOI and pDOI has been well described [31] and is consistent with the current data; thus, the AJCC 8th edition includes it as a parameter for T staging (≤5 mm and >5 mm; ≤10 mm and >10 mm) [6,14,42,43]. Most ENEs were considered to have occurred in the advanced stage (93%, 25/27) with DOI >10 mm, and 79% of cases with lymph node metastasis (46 of 58) had a DOI >10 mm; thus, DOI >10 mm could increase the sensitivity and/or speci city of factors, such as DR-I/TILs-L and TB-H.
Our study demonstrated the association between DR-I and higher pT and the presence of lymph node metastasis, including ENE. Previous studies have revealed that DR-I is associated with higher pT, presence of lymph node metastasis, and poor outcome in OSCC [17,21,23]. This might be due to the transformation of broblasts to cancer-associated broblasts (CAFs) during carcinogenesis [24]. CAFs promote tumor invasiveness and remodel the stroma from a mature to immature state by the deposition of glycosaminoglycans through growth differentiation factor (GDF10) secretion [44,45]. This represents the histopathological features of DR-I [24,44]. Therefore, the evaluation of DR in both biopsies and resections is a useful predictive factor for tumor behavior and lymph node metastasis in patients with OSCC.
Our results showed that TILs-L was signi cantly associated with DOI >5 mm, DOI >10 mm, lymph node metastasis, and the number of metastatic nodes in resections and biopsies. TILs are a selected population of T-cells with a higher speci c immunological reactivity against tumor cells [17,18].
In OSCC, TILs also include B-cells, and B-cell density is a good prognostic predictor in male and younger patients with OSCC who consume tobacco or alcohol [46]. TILs and a modulator of cancer invasion and metastasis [17,18] and low-grade lymphocyte in ltration is associated with worse outcomes [19]; these reports are consistent with the current results.
In this study, a single pathological risk factor showed high sensitivity but low speci city and vice versa. In biopsies, the combination of DR-I/TILs-L/cDOI >10 mm successfully had higher sensitivity, speci city, and accuracy (70%, 77%, and 76%, respectively), whereas TB-H/pDOI >10 mm exhibited higher speci city and accuracy (91% and 83%, respectively) than a single factor in resections.
These predictive factors could be simply evaluated using routine H&E staining and included in pathology reports. Few studies have examined the pathological predictive factors for ENE [40,41]. Gleber-Netto et al. noted an increasing trend toward the frequency of ENE among patients with TP53mutant OSCC (65.5%) than among those with WT TP53 OSCC (42.1%; p=0.07) [40]. Dhanda et al. showed that the combination of smooth muscle actin and SERPINE1 expression (4G/5G polymorphism) in CAF had superior sensitivity and speci city to MRI for the detection of ENE (sensitivity: 81%; speci city: 54%) [41]. Predictors from biopsies, such as DR-I/TILs-L, also had high sensitivity and accuracy (80% and 70%, p<0.01), and adding cDOI >10 mm (DR-I/TILs-L/cDOI>10 mm) provided high sensitivity, speci city, and accuracy (70%, 55%, and 76%, respectively; p<0.01); this could also be evaluated simply even in biopsies. In addition, histological features, such as non-TB-H/pDOI >10 mm, in resections were revealed to be excellent negative predictors of ENE, with high speci city and accuracy. These two predictors could also predict all ENEs, including ENEmi.
This study has some limitations. First, the sample size was relatively small, particularly in patients with ENE. Second, we did not perform survival analysis, and the prognostic value derived with the available predictors was unclear, although the presence of ENE is an established prognostic factor in patients with OSCC.

Conclusions
Our study showed that TB-H was an independent risk factor for ENE because the TME status in primary OSCC was signi cantly associated with that in ENE. This is the rst study to demonstrate the association among DR, TB, TILs, and DOI between primary OSCC and ENE.