pT1 结直肠癌淋巴结转移风险：「瘤周恶性细胞灶」优于肿瘤出芽与低分化细胞簇Peritumoral Malignant Cellular Foci Performs Better Than Tumor Budding and Poorly Differentiated Cluster in Risk Stratification of Lymph Node Metastasis in pT1 Colorectal Cancer

Scope: open-access full text (CC BY-NC-ND 4.0, Ovid/LWW) — abstract, introduction, materials and methods, results, discussion and acknowledgments, with 6 figures and 7 tables (including footnotes); references and supplemental digital content are not included. Superscript numbers are the original reference numbers.

Abstract

As biomarkers separated by the arbitrary 5-tumor-cell cutoff at the tumor invasive front, tumor budding (TB) and poorly differentiated clusters (PDC) are recognized as high-risk factors for lymph node metastasis (LNM) in pT1 colorectal cancer (CRC). Nonetheless, relevant findings remain inconsistent. Furthermore, the association between the tumor invasive front biomarker and D3 LNM in pT1 CRC remains unknown. To identify a biomarker that captures the full morphologic spectrum and eliminates the need for such cutoffs at the tumor invasive front for predicting LNM, particularly D3 LNM, in pT1 CRC, we defined peritumoral malignant cellular foci (PMCF) as nonglandular foci composed of ≥1 tumor cell at the tumor invasive front. After enrolling 2611 endoscopically or surgically resected pT1 CRC patients from 6 high-volume Chinese institutions, we found that neither TB nor PDC exhibited a consistent independent association with LNM across the training and validation cohorts. In contrast, PMCF emerged as an independent risk factor with superior risk stratification of LNM in both cohorts (P<0.05). The association between PMCF and LNM remained significant in subgroup analyses stratified by age, sex, tumor location, and BMI (P<0.05). Moreover, PMCF correlated with a greater number of metastatic lymph nodes and exhibited a stronger association with D3 LNM than with D1 and D2 LNM (P<0.05). In conclusion, PMCF not only suggests additional radical surgery after endoscopic resection in pT1 CRC, but may also indicate D3 lymph node dissection. PMCF provides an evidence-based rationale to optimize surgical decision-making in pT1 CRC.

Introduction

Tumor budding (TB) and poorly differentiated cluster (PDC) are typical biomarkers at the tumor invasive front. Since lymph node metastasis (LNM) occurs in only 10% to 15% of patients with pT1 colorectal cancer (CRC),¹ current guidelines recognize high-grade TB as a risk factor for LNM and recommend additional radical surgery (RS) after endoscopic resection (ER) to achieve risk stratification of LNM in pT1 CRC.^2–4 However, emerging evidence suggested that TB lost its association with LNM after adjustment for PDC.^5,6 Although PDC showed potential for predicting LNM in pT1 CRC,^6,7 the 5-tumor-cell cutoff that artificially separates PDC and TB fails to reflect the continuous morphologic spectrum as well as the biological continuum of the tumor invasive front and impedes their clinical application.^8,9 Due to their extensive overlapping molecular characteristics^10,11 combined assessment of TB and PDC in CRC would yield better prognostic performance than either alone.^12,13 Haddad et al⁸ reported that a threshold of ≤5 cells for defining tumor-cluster size yielded the optimal stratification for disease-free survival in stage I to III CRC patients.⁸ Comprehensive evaluation of the morphologic features at the tumor invasive front is therefore essential. However, studies on a biomarker that captures the full morphologic spectrum of the tumor invasive front for predicting LNM and guiding additional RS in pT1 CRC still remain limited.

Numerous studies have demonstrated that the main LNM (D3 LNM) is an indicator of poor prognosis in CRC.^14,15 Accurate identification of D3 LNM is therefore significant for determining the extent of lymph node dissection in pT1 CRC patients. However, risk factors for D3 LNM in pT1 CRC remain largely unknown, particularly the association between the tumor invasive front biomarker and D3 LNM. Although the incidence of D3 LNM in pT1 CRC is 0.3%,² D3 lymph node dissections might not be indicated for a proportion of patients. Consequently, these critical knowledge gaps have left surgeons without sufficient evidence to determine the necessity of D3 lymph node dissection and identify pT1 CRC patients most likely to benefit from D3 dissection. Thus, refining the evidence-based rationale for surgical decision-making in pT1 CRC is critically imperative.

To address these clinical dilemmas and knowledge gaps, we proposed and defined the peritumoral malignant cellular foci (PMCF) as nonglandular foci composed of ≥1 tumor cell at the tumor invasive front. We therefore aimed to evaluate the predictive performance of PMCF for LNM in pT1 CRC and explore its association with D3 LNM.

Materials and Methods

Study Cohort

Clinical data of pT1 CRC patients were collected from 6 high-volume Chinese institutions. Patients who underwent ER alone, initial or additional RS after ER with pathologically confirmed pT1 CRC were included. For patients treated with ER alone and additional RS after ER, endoscopic specimens and surgical specimens were used for pathologic assessment. For patients treated with initial RS, surgical specimens were used for pathologic assessment. According to the Japanese Society for Cancer of the Colon and Rectum (JSCCR) guidelines,² patients who had one or more risk factors outlined below were recommended for additional RS with lymph nodes dissection after ER: (1) positive lymphovascular invasion (LVI); (2) poorly differentiated adenocarcinoma, signet ring cell carcinoma or mucinous carcinoma; (3) TB grade of BD 2/3; (4) depth of submucosal invasion ≥1000 µm; and (5) positive vertical margin. Due to physical condition and personal preference, some of the patients refused additional RS. Patients who had (1) synchronous or metachronous CRC, (2) inflammatory bowel disease, (3) familial adenomatous polyposis or Peutz–Jeghers syndrome, (4) malignancy from other organs, (5) neoadjuvant chemotherapy or radiotherapy, (6) distant metastasis at diagnosis, (7) <5 years follow-up after ER alone, and (8) missing clinical data were excluded. A total of 2611 patients were enrolled. One thousand one hundred eight pT1 CRC patients enrolled from the Sixth Affiliated Hospital of Sun Yat-sen University (February 2010 to May 2025) served as the training cohort. A total of 1503 patients were enrolled as the validation cohort, including 473 patients from Sun Yat-sen University Cancer Center (April 2015 to August 2024), 111 from Shaoguan Yuebei People’s Hospital (January 2014 to November 2024), 270 from the Third Bethune Hospital of Jilin University (July 2017 to May 2025), 263 from the Second Affiliated Hospital of Harbin Medical University (April 2018 to May 2025), and 208 from Shenzhen People’s Hospital (January 2014 to November 2024). In addition, 178 patients prospectively enrolled from the Sixth Affiliated Hospital of Sun Yat-sen University (June 2025 to January 2026) with the same criteria were recruited in the validation cohort. This purely observational study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Ethics Committee of the Sixth Affiliated Hospital of Sun Yat-sen University (No. 2025ZSLYEC-299) as well as other participating centers. Written informed consent was signed by each participant at their hospital admission.

Sample Size Calculation

The sample size was estimated before the study based on the method of events per variable (EPV).¹⁶ Based on the classic and widely adopted EPV of 10,¹⁷ with an assumed LNM-positive rate of 10% to 15%¹ and 22 candidate variables, the required range of the sample size was calculated as follows:

The Range of Sample Size

=

EPV

×

Number of Variables

Incidence Rate

=

10

×

22

0.10

~

0.15

=

1

,

467

~

2200

The total cohort of 2611 patients substantially exceeded this range, ensuring an adequate sample size for analysis.

Histologic Assessment and Clinical Data Collection

All hematoxylin and eosin (H&E)–stained and immunohistochemistry (IHC) stained tumor slides were retrieved from institutional pathology archives. All histologic parameters were assessed using longitudinal sections of the whole tumor, including its deepest invasive part. For rigorous pathologic assessment, all the tumor slides (n=2611, 13,881 slides) were independently reviewed by 2 experienced gastrointestinal pathologists (X.L. and Q.Z.) who were blinded to clinical outcomes. In cases of discordance on initial review, a consensus was reached via review by a third pathologist (C.W.) and joint discussion.

TB and PDCs

TB was defined as an isolated single tumor cell or a cluster comprised of <5 tumor cells without glandular structures at the tumor invasive front.¹⁸ Briefly, the invasive front of all tumor slides was scanned at low-power magnification to identify the area with the greatest density of TBs. After the identification of the hotspot area, the number of buds was counted under a ×20 objective lens (0.785 mm²). In accordance with the International Tumor Budding Consensus Conference (ITBCC) recommendations, the TB was graded as grade 1 (0 to 4 buds), grade 2 (5 to 9 buds), or grade 3 (≥10 buds).¹⁹ According to the JSCCR guidelines,² grade 1 was classified as low grade (Fig. 1A) and grades 2 and 3 were classified as high grade (Fig. 1B).

PDC was defined as clusters of ≥5 tumor cells with no evidence of glandular differentiation at the tumor invasive front.²⁰ Similarly, the invasive front of all the tumor slides was carefully scanned to identify the area with the greatest density of PDC. After localizing the hotspot area, PDC was counted and classified into negative (Fig. 1C) or positive (Fig. 1D) based on its presence under a ×20 objective lens (0.785 mm²).⁷

Peritumoral Malignant Cellular Foci

We defined PMCF as nonglandular foci composed of ≥1 tumor cell at the tumor invasive front. First, tumor invasive margin in all the tumor slides was assessed to identify the area with the greatest density of PMCF. The count of PMCF was performed under a ×20 objective lens within the hotspot area (0.785 mm²). Since PMCF was identified as an independent LNM risk factor in both the training and validation cohorts (Supplementary Table S1, Supplemental Digital Content 1, Supplemental Digital Content 1, https://links.lww.com/PAS/C439), the optimal cutoff was determined by the Youden index in the training cohort at 3 (Supplementary Figure S1, Supplemental Digital Content 2, Supplemental Digital Content 1, https://links.lww.com/PAS/C439). Tumors with <3 PMCF were categorized as low grade (Fig. 1E) and those with ≥3 PMCF as high grade (Fig. 1F). Specifically, any focus defined as a PMCF was also counted as either a TB or a PDC.

Other Clinical Data and Histologic Parameters

Clinical and demographical information, including age, sex, body mass index (BMI), tumor location, tumor size, family history of CRC, preoperative contrast-enhanced CT reports, colonoscopy images, and serum tumor marker reports, were retrieved from the electronic medical record system.

The tumor grade (TG) was classified into well-differentiated, moderately differentiated, poorly differentiated adenocarcinoma, mucinous carcinoma, and signet ring cell carcinoma according to the WHO criteria.²¹ For tumors containing more than one component, the grade was determined based on the least differentiated component. Ki-67 expression was assessed as the percentage of positively stained tumor cells by IHC staining. Since continuous variables including Ki-67 and preoperative carcinoembryonic antigen (pre-CEA) were also identified as independent LNM risk factors in both the training and validation cohorts (Supplementary Table S1, Supplemental Digital Content 1, https://links.lww.com/PAS/C439), their optimal cutoffs were determined by ROC in the training cohort at 75% and 4.90, respectively (Supplementary Figure S1, Supplemental Digital Content 2, Supplemental Digital Content 1, https://links.lww.com/PAS/C439). LVI was defined as the presence of tumor cells within lymphatic or blood vessels.²² According to the NCCN guidelines^3,4 and the US Multi-Society Task Force guidelines,²³ when LVI was equivocal on H&E staining, IHC staining of anti-D2-40 antibody and anti-CD34 antibody was performed. Perineural invasion (PNI) was defined as the nerve sheath encircled by tumor cells.²⁴ The depth of submucosal invasion was measured in accordance with the JSCCR guideline.² In brief, upon the identification or estimation of the muscularis mucosae (MM), invasion depth was measured from its lower border to the deepest invasion. When the MM was unable to be identified or estimated, invasion depth was measured from the tumor surface to the deepest invasive front. For pedunculated lesions with tangled MM, the junction between the tumor head and neck (Haggitt level 2) was utilized as the reference line. Invasion confined to the head was recorded as 0 μm (head invasion), whereas invasion extending beyond (stalk invasion) was measured from the reference line to the deepest invasive front. MM status was evaluated with the assistance of antidesmin antibody and classified as Grade 1 when muscle fibers remained identifiable or Grade 2 when they were severely fragmented or entirely absent.²⁵ Background adenoma (BGA) was defined as an adenomatous component contiguous with the carcinoma.²⁶ IHC staining results of HER-2 were evaluated according to the American Society of Clinical Oncology (ASCO) guidelines.^27,28 The expression of mismatch repaired (MMR) proteins (MLH1, PSM2, MSH2, and MSH6) were determined by IHC staining. Specimens with intact staining for all MMR proteins were classified as proficient MMR (pMMR). Tumors lacking intact MMR proteins expression were categorized as deficient MMR (dMMR). The IHC technical specifications for D2-40, CD34, HER2, Ki67, and MMR were listed in Supplemental Digital Content 3, Supplemental Digital Content 1, https://links.lww.com/PAS/C439.

Tumor location was categorized into the right colon (cecum, ascending colon, and transverse colon) and the left colon and rectum (descending colon, sigmoid colon, and rectum). The maximum tumor diameter was measured after formalin fixation. Tumor morphology was assessed according to the Paris endoscopic classification²⁹ and dichotomized into the depressed type (IIc, III, Is+IIc, IIa+IIc, or IIc+IIa) and the nondepressed type (Ip, Isp, Is, IIa, or IIb). The BMI was classified into nonoverweight (BMI <24 kg/m²) and overweight (BMI ≥24 kg/m²). Patients with CRC diagnosed before the age of 50 were classified as early-onset (EOCRC), whereas those aged 50 or older were classified as late-onset (LOCRC).

Assessment of Study Outcomes

For patients who underwent initial or additional RS, LNM was determined via pathologic examination of resected lymph node specimens. For patients treated with ER alone, a follow-up of ≥5 years is needed in accordance with the JSCCR guidelines.² LNM was evaluated via contrast-enhanced CT scan.^30,31 Patients were defined as LNM-negative if the results of contrast-enhanced CT scans were free from regional lymph node (LN) recurrence for 5 years or longer, whereas those with radiologic evidence of regional LN recurrence at any time during follow-up were classified as LNM-positive.

Diagnosis of regional LN recurrence using contrast-enhanced CT was performed in accordance with the following criteria: (1) maximum short-axis diameter >5 mm; (2) abnormal nodal density or enhancement; (3) nodal necrosis or obscure borders; and (4) matted lymph nodes or nodal fusion with adjacent organs or vessels. Lesions presenting any of the above features were defined as positive LN recurrence. All the participating institutions applied the same criteria.

Regional LN stations were classified according to the Japanese Classification of Colorectal Carcinoma.³² Briefly, first-station LN (D1) were defined as pericolic/perirectal nodes, second-station LN (D2) as intermediate nodes, and third-station LN (D3) as main nodes. Skip metastasis was defined as the presence of tumor metastasis in the distant regional lymph node stations without involvement of the proximal stations. Otherwise, it was defined as nonskip metastasis.

Developing an LNM Prediction Model

In the training cohort, bidirectional stepwise regression was performed to screen 20 candidate variables, with TB and PDC excluded from variable selection. Fourteen key predictors—LVI (positive vs. negative), PNI (positive vs. negative), Pre-CEA (ng/mL), BGA (present vs. absent), PMCF (low-grade vs. high-grade), tumor morphology (depressed vs. nondepressed), TG (well-/moderate differentiation vs. poor differentiation/signet ring cell/mucinous carcinoma), HER-2 (−/1+ vs. 2+/3+), Ki-67 (%), family history of CRC (yes vs. no), tumor location (right colon vs. left colon/rectum), MM status (Grade 1 vs. Grade 2), tumor size (mm), and CT examination (positive vs. negative)—were subsequently retained to construct a logistic regression (LR) model.

The risk score was calculated using the following formula: −5.6218+1.75374*LVI +0.14447*Pre-CEA +0.02534*Ki-67-0.85757*BGA +0.80265*PMCF +0.75279*CT Examination+1.11027*HER-2+1.1447*TG+0.81415*Tumor Morphology+0.70308*MM status +1.62325*PNI −1.20887*Family History of CRC+0.59778*Tumor Location-0.01561*Tumor Size. The optimal cutoff value of −2.3583 was determined by the Youden index. The identical formula and cutoff were then applied to the validation cohort. The predictive performance of the model was comprehensively compared with current guidelines using receiver operating characteristic (ROC), precision-recall curves (PR), and decision curve analysis (DCA) in the training and validation cohorts. To facilitate its clinical application, we developed a web calculator.

According to the JSCCR guidelines,² patients who have one or more risk factors outlined below are defined as high risk of LNM and recommended for additional RS: (1) positive LVI; (2) poorly differentiated adenocarcinoma, signet ring cell carcinoma, or mucinous carcinoma; (3) TB grade of BD 2/3; (4) depth of submucosal invasion ≥1000 µm. The risk factor of high-grade TB is absent from the ESMO guidelines,^33,34 while the depth of invasion is not included in the NCCN guidelines.^3,4

Interobserver Agreement

We randomly selected the original records of 50 cases that had been independently reviewed previously by the 2 aforementioned pathologists (X.L. and Q.Z.) and performed the interobserver agreement of TB, PDC, and PMCF.

Statistical Analysis

Continuous variables were presented as mean±SD and compared using the t test or the Wilcoxon test depended on their distributions. Categorical variables were presented as frequencies and percentages. Independent risk factors were identified using multivariate LR. To evaluate whether the effects of risk factors differed significantly across subgroups stratified by age, sex, tumor location, and BMI, interaction tests (P interaction) were performed. Likelihood ratio tests with one degree of freedom (P heterogeneity) were conducted to evaluate the heterogeneity of risk factor effects across LNM subgroups in multinomial LR. Briefly, this test compared a model that allows for separate associations of risk factors with LNM according to metastasis subtypes with a model that assumes a common association.³⁵ Predictive performance was assessed using ROC with DeLong tests and PR. DCA was performed to quantify the net benefit across a range of threshold probabilities. Cohen Kappa coefficients were calculated to quantify the interobserver agreement. A 2-tailed P-value <0.05 was considered statistically significant. All statistical analyses were performed with R (Version 4.4.0).

Results

Baseline Clinical Information

One hundred fifteen (10.4%) patients were LNM-positive in the training cohort, whereas 168 (11.2%) were LNM-positive in the validation cohort. Overall, the LNM-positive rate was 10.8%. Demographic and clinical characteristics of all patients were summarized in Table 1. The patients flowchart was illustrated in Figure 2.

BGA indicates background adenoma; BMI, body mass index; CRC, colorectal cancer; dMMR, deficient mismatch repair; LN, lymph node; LNM, lymph node metastasis; LVI, lymphovascular invasion; MMR, mismatch repair; Muc, mucinous carcinoma; PDC, poorly differentiated cluster; PMCF, peritumoral malignant cellular foci; pMMR, proficient mismatch repair; PNI, perineural invasion; Pre-CA199, preoperative CA199; Pre-CEA, preoperative carcinoembryonic antigen; Sig, Signet ring cell carcinoma; TB, tumor budding.

High-Grade PMCF was an Independent Risk Factor for LNM With Superior Risk Stratification

We systematically compared the predictive performance of TB, PDC, and PMCF for LNM risk stratification in pT1 CRC. Notably, Grade 2/3 TB was not an independent high-risk factor in either the training or the validation cohort (P>0.05, Table 2). While positive PDC was identified as an independent risk factor in the validation cohort (P<0.05); it failed to do so in the training cohort (P>0.05, Table 2). By contrast, high-grade PMCF was identified as an independent risk factor in both cohorts (P<0.05, Table 2, Supplementary Table S2, Supplemental Digital Content 4, Supplemental Digital Content 1, https://links.lww.com/PAS/C439). In addition, as illustrated in Figure 3, PMCF exhibited the strongest LNM risk stratification in both the training and validation cohorts (P<0.05).

^* Continuous variables including age, pre-CA199, BMI, tumor size, and categorical variables including sex (male/female), tumor location (right colon/left colon and rectum), LVI (negative/positive), PNI (negative/positive), TG (well and moderate /poor and muc and sig), depth of invasion (<1000 µm/≥1000 µm), morphology (undepressed/depressed), tumor budding (Grade 1/Grade 2 and 3), PDC (negative/positive), MM (Grade 1/Grade 2), BGA (presence/absence), family history of CRC (yes/no), MMR (dMMR/pMMR), HER-2 (- and 1+/2+ and 3+), Ki-67 (<75%/≥75%), Pre-CEA (＜4.90/≥4.90), CT examination (negative/positive) were adjusted in multivariate logistic regression.

^† Continuous variables including age, pre-CA199, BMI, tumor size and categorical variables including sex (male/female), tumor location (right colon/left colon and rectum), LVI (negative/positive), PNI (negative/positive), TG (well and moderate /poor and muc and sig), depth of invasion (<1000 µm/≥1000 µm), morphology (undepressed/depressed), PMCF (low grade/high grade), MM (Grade 1/Grade 2), BGA (presence/absence), family history of CRC (yes/no), MMR (dMMR/pMMR), HER-2 (- and 1+/2+ and 3+), Ki-67 (<75%/≥75%), Pre-CEA (＜4.90/≥4.90), CT examination (negative/positive) were adjusted in multivariate logistic regression.

BGA indicates background adenoma; BMI, body mass index; CEA, carcinoembryonic antigen; CRC, colorectal cancer; dMMR, deficient mismatch repair; LNM, lymph node metastasis; LVI, lymphovascular invasion; MMR, mismatch repair; Muc, mucinous carcinoma; PDC, Poorly differentiated cluster; PMCF, peritumoral malignant cellular foci; pMMR, proficient mismatch repair; PNI, perineural invasion; Sig, Signet ring cell carcinoma; TB, tumor budding.

High-Grade PMCF was a Robust Predictor of LNM in Various Subgroups

To further compare the LNM predictive performance of TB, PDC, and PMCF across subgroups stratified by age, sex, tumor location, and BMI, subgroup analysis with an interaction test was conducted. High-grade PMCF was identified as an independent risk factor for LNM across all subgroups (P<0.05, Table 3). Moreover, high-grade PMCF was associated with a significantly higher LNM risk in right colon tumors compared with left colon and rectal tumors (P interaction <0.05, Table 3). In contrast, Grade 2/3 TB did not exhibit independent associations with LNM in any subgroup analysis (P>0.05, Table 3). Subgroup analyses revealed that positive PDC was an independent risk factor for LNM in the following subgroups: LOCRC, right-sided CRC, left-sided CRC, male patients, normal-weight patients, and overweight patients (P<0.05, Table 3). Detailed information of subgroup analysis was listed in Supplementary Table S3-S6 (Supplemental Digital Content 5-8, Supplemental Digital Content 1, https://links.lww.com/PAS/C439).

^* The OR and P-values were derived from multivariate logistic regression. Continuous variables including age, pre-CA199, BMI, tumor size, and categorical variables including sex (male/female), tumor location (right colon/left colon and rectum), LVI (negative/positive), PNI (negative/positive), TG (well and moderate /poor and muc and sig), depth of invasion (<1000 µm/≥1000 µm), morphology (undepressed/depressed), tumor budding (Grade 1/Grade 2 and 3), PDC (negative/positive), MM (Grade 1/Grade 2), BGA (presence/absence), family history of CRC (yes/no), MMR (dMMR/pMMR), HER-2 (-&1+/2+&3+), Ki-67 (<75%/≥75%), pre-CEA (＜4.90/≥4.90), CT examination (negative/positive) were adjusted in multivariate logistic regression. P for Interaction was calculated through the interaction test.

^† The OR and P-values were derived from multivariate logistic regression. Continuous variables including age, pre-CA199, tumor size, BMI and categorical variables including sex (male/female), tumor location (right colon/left colon and rectum), LVI (negative/positive), PNI (negative/positive), TG (well and moderate /poor and muc and sig), depth of invasion (<1000 µm/≥1000 µm), morphology (undepressed/depressed), PMCF (low grade/high grade), MM (Grade 1/Grade 2), BGA (presence/absence), family history of CRC (yes/no), MMR (dMMR/pMMR), HER-2 (-&1+/2+&3+), Ki-67 (<75%/≥75%), pre-CEA (＜4.90/≥4.90), CT examination (negative/positive) were adjusted in multivariate logistic regression. P for Interaction was calculated through the interaction test.

BGA indicates background adenoma; BMI, body mass index; CEA, carcinoembryonic antigen; CRC, colorectal cancer; dMMR, deficient mismatch repair; EOCRC, early onset colorectal cancer; LNM, lymph node metastasis; LOCRC, late onset colorectal cancer; LVI, lymphovascular invasion; MMR, mismatch repair; Muc, mucinous carcinoma; PDC, poorly differentiated clusters; PMCF, peritumoral malignant cellular foci; pMMR, proficient mismatch repair; PNI, perineural invasion; Sig, Signet ring cell carcinoma; TB, tumor budding.

High-Grade PMCF was Associated With a Greater Number of Metastatic Lymph Nodes

After excluding 42 patients who underwent ER only, there were 274 LNM-positive patients in the total remaining cohort. We found that high-grade PMCF was associated with a significantly higher number of metastatic lymph nodes (P<0.05, Fig. 4A). However, neither Grade 2/3 TB nor positive PDC was associated with the number of metastatic lymph nodes (P>0.05, Fig. 4B-C). In addition, as independent risk factors in both the training and validation cohorts (Supplementary Table S2, Supplemental Digital Content 4, Supplemental Digital Content 1, https://links.lww.com/PAS/C439), absent BGA, positive CT examination, HER-2 with 2+/3+, Ki-67 ≥75% and positive LVI were also associated with a higher number of metastatic lymph nodes (P<0.05, Supplementary Fig. S2, Supplemental Digital Content 9, Supplemental Digital Content 1, https://links.lww.com/PAS/C439). The number of retrieved lymph nodes was comparable across those histologic parameters among these 274 LNM-positive patients (P>0.05, Supplementary Fig S3, Supplemental Digital Content 10, Supplemental Digital Content 1, https://links.lww.com/PAS/C439).

High-Grade PMCF was an Independent Risk Factor for D3 LNM

To investigate the predictive performance of PMCF for D3 LNM, we excluded patients who underwent ER alone, as well as LNM-positive cases without a specified lymph node station, from the total cohort. Multinomial LR and the heterogeneity test were then performed on the remaining 2485 patients. As shown in Table 4, high-grade PMCF was an independent risk factor for D3 LNM (P<0.05). Furthermore, it exhibited a much stronger association with D3 LNM than with D1 and D2 LNM (P heterogeneity <0.05). Similarly, high-grade PMCF was identified as an independent risk factor for skip metastasis (P<0.05, Table 5). In addition, positive PDC was identified as an independent risk factor for D1 and D2 LNM as well as nonskip metastasis (P<0.05, Table 4 and Table 5). However, neither TB nor PDC was an independent risk factor for D3 and skip metastasis (P>0.05, Table 4 and Table 5).

^* pT1 CRC patients who underwent ER alone, as well as LNM-positive cases with unspecified lymph node station, were excluded from the total cohort.

^† The OR and P-values were derived from multivariate logistic regression. Continuous variables including age, pre-CA199, BMI, tumor size, and categorical variables including sex (male/female), tumor location (right colon/left colon and rectum), LVI (negative/positive), PNI (negative/positive), TG (well and moderate /poor and muc and sig), depth of invasion (<1000 µm/≥1000 µm), morphology (undepressed/depressed), tumor budding (Grade 1/Grade 2 and 3), PDC (negative/positive), MM (Grade 1/Grade 2), BGA (presence/absence), family history of CRC (yes/no), MMR (dMMR/pMMR), HER-2 (-&1+/2+&3+), Ki-67 (<75%/≥75%), Pre-CEA (＜4.90/≥4.90), CT examination (negative/positive) were adjusted in multivariate logistic regression. P for heterogeneity was calculated through a likelihood ratio test with one degree of freedom.

^‡ The OR and P-values were derived from multivariate logistic regression. Continuous variables including age, pre-CA199, tumor size, BMI and categorical variables including sex (male/female), tumor location (right colon/left colon and rectum), LVI (negative/positive), PNI (negative/positive), TG (well and moderate /poor and muc and sig), depth of invasion (<1000 µm/≥1000 µm), morphology (undepressed/depressed), PMCF (low grade/high grade), MM (Grade 1/Grade 2), BGA (presence/absence), family history of CRC (yes/no), MMR (dMMR/pMMR), HER-2 (-&1+/2+&3+), Ki-67 (<75%/≥75%), pre-CEA (＜4.90/≥4.90), and CT examination (negative/positive) were adjusted in multivariate logistic regression. P for heterogeneity was calculated through a likelihood ratio test with one degree of freedom.

BGA indicates background adenoma; BMI, body mass index; CEA, carcinoembryonic antigen; CRC, colorectal cancer; D1, pericolic/perirectal lymph nodes; D2, intermediate lymph nodes; D3, main lymph nodes; dMMR, deficient mismatch repair; LNM, lymph node metastasis; LVI, lymphovascular invasion; MMR, mismatch repair; Muc, mucinous carcinoma; PDC, poorly differentiated clusters; PMCF, peritumoral malignant cellular foci; pMMR, proficient mismatch repair; PNI, perineural invasion; Sig, Signet ring cell carcinoma; TB, tumor budding.

^* pT1 CRC patients who underwent ER alone, as well as LNM-positive cases with unspecified lymph node station, were excluded from the total cohort.

^† The OR and P-values were derived from multivariate logistic regression. Continuous variables including age, pre-CA199, BMI, tumor size, and categorical variables including sex (male/female), tumor location (right colon/left colon and rectum), LVI (negative/positive), PNI (negative/positive), TG (well and moderate /poor and muc and sig), depth of invasion (<1000 µm/≥1000 µm), morphology (undepressed/depressed), tumor budding (Grade 1/Grade 2 and 3), PDC (negative/positive), MM (Grade 1/Grade 2), BGA (presence/absence), family history of CRC (yes/no), MMR (dMMR/pMMR), HER-2 (-&1+/2+&3+), Ki-67 (<75%/≥75%), Pre-CEA (＜4.90/≥4.90), CT examination (negative/positive) were adjusted in multivariate logistic regression. P for heterogeneity was calculated through a likelihood ratio test with one degree of freedom.

^‡ The OR and P-values were derived from multivariate logistic regression. Continuous variables including age, pre-CA199, tumor size, BMI, and categorical variables including sex (male/female), tumor location (right colon/left colon and rectum), LVI (negative/positive), PNI (negative/positive), TG (well and moderate /poor and muc and sig), depth of invasion (<1000 µm/≥1000 µm), morphology (undepressed/depressed), PMCF (low grade/high grade), MM (Grade 1/Grade 2), BGA (presence/absence), family history of CRC (yes/no), MMR (dMMR/pMMR), HER-2 (-&1+/2+&3+), Ki-67 (<75%/≥75%), Pre-CEA (＜4.90/≥4.90), CT examination (negative/positive) were adjusted in multivariate logistic regression. P for heterogeneity was calculated through a likelihood ratio test with one degree of freedom.

BGA indicates background adenoma; BMI, body mass index; CEA, carcinoembryonic antigen; CRC, colorectal cancer; dMMR, deficient mismatch repair; LNM, lymph node metastasis; LVI, lymphovascular invasion; MMR, mismatch repair; Muc, mucinous carcinoma; PDC, poorly differentiated clusters; PMCF, peritumoral malignant cellular foci; pMMR, proficient mismatch repair; PNI, perineural invasion; Sig, Signet ring cell carcinoma; TB, tumor budding.

In addition, the heterogeneity test revealed that positive LVI, depressed tumor morphology, absent BGA, and Ki-67 ≥75% exhibited stronger associations with D3 LNM (P heterogeneity <0.05, Supplementary Table S7, Supplemental Digital Content 11, Supplemental Digital Content 1, https://links.lww.com/PAS/C439). However, no statistically significant differences were found between skip metastasis and nonskip metastasis (P heterogeneity >0.05, Supplementary Table S8, Supplemental Digital Content 12, Supplemental Digital Content 1, https://links.lww.com/PAS/C439).

Associations Between PMCF and Other Histologic Parameters

As shown in Table 6, high-grade PMCF was significantly correlated with positive LVI, positive PNI, poor tumor differentiation, smaller tumor size, deeper submucosal invasion depth, depressed tumor morphology, severe or complete loss of MM, absence of BGA, elevated Ki-67 expression, and pre-CEA expression (P<0.05).

BGA indicates background adenoma; BMI, body mass index; CRC, colorectal cancer; dMMR, deficient mismatch repair; LVI, lymphovascular invasion; MMR, mismatch repair; Muc, mucinous carcinoma; PMCF, peritumoral malignant cellular foci; pMMR, proficient mismatch repair; PNI, perineural invasion; Pre-CA199, preoperative CA199; Pre-CEA, preoperative carcinoembryonic antigen; Sig, Signet ring cell carcinoma.

Development of a PMCF-Based LNM Prediction Model

A PMCF-based LNM prediction model was constructed using the LR algorithm. As illustrated in Figure 5, this model exhibited significantly better predictive performance than current clinical guidelines (P<0.05) in both the training and validation cohorts. Therefore, the PMCF-based LNM prediction model holds potential to guide additional RS. In addition, to facilitate the clinical use of this model, we have constructed a web calculator (Fig. 6, Supplemental Digital Content 13, Supplemental Digital Content 1, https://links.lww.com/PAS/C439).

Interobserver Agreement

PMCF yielded substantial interobserver agreement (Cohen Kappa=0.700, P<0.05), a value similar to the Kappa of 0.719 for PDC and far higher than that for TB (Cohen Kappa=0.563, Table 7).

PDC indicates poorly differentiated clusters; PMCF, peritumoral malignant cellular foci; TB, tumor budding.

Discussion

In this study, we found that PMCF provides better LNM risk stratification than either TB or PDC in pT1 CRC. The association between PMCF and LNM remained significant in subgroup analyses stratified by age, sex, tumor location, and BMI. In addition, high-grade PMCF was associated with a significantly greater number of metastatic lymph nodes. Furthermore, as an emerging risk factor for D3 LNM, high-grade PMCF showed a markedly stronger association with D3 LNM than with D1 and D2 LNM.

In current guidelines, additional RS is recommended when Grade 2/3 TB is identified.^2–4 However, in our results, TB ceased to be an independent risk factor upon adjustment for PDC. This finding aligns with previous studies,^5,6 which similarly demonstrated that TB was no longer an independent predictor following adjustment for PDC. Although PDC has not yet been incorporated into any clinical guidelines, emerging evidence suggests its potential as a predictive marker.^6,7 Notably, in our results, after adjustment for TB, PDC was not a robust independent risk factor. We postulated that the predictive performance of TB and PDC for LNM in pT1 CRC is substantially influenced by their mutual interaction. Researchers have found considerable overlapping between TB and PDC in terms of molecular characteristics and clinical utility. Compared with the background tumor, both TB and PDC exhibited elevated nuclear β-Catenin and reduced E-Cadherin expression.¹⁰ KRAS mutations were more frequent in both high-grade TB and PDC.³⁶ TB and PDC engaged in dynamic epithelial-mesenchymal transition (EMT).^37,38 A recent research found that MYC downregulation and RUNX2 upregulation contributed to the acquisition of an invasive phenotype with TB and PDC.³⁹ Moreover, the combination of TB and PDC provides superior prognostic stratification in CRC.⁴⁰ Based on our findings and previous research, it is plausible that TB and PDC represent a continuous invasive spectrum with shared biological features as one pathologic entity. Regardless of cellularity within each tumor cell cluster at the invasive front, a greater number of clusters correlates with more aggressive LNM progression. Therefore, a separate assessment of TB or PDC might not comprehensively reflect the biological continuum of pT1 CRC LNM. This prompts critical questions: Is the 5-tumor-cell threshold a reasonable cutoff? May TB and PDC be considered as one integrated pathologic entity? Multiple studies have investigated the combination of TB and PDC for predicting LNM in pT1 CRC. Backes et al⁵ proposed that the presence of either TB of Grade 2/3 or PDC of Grade 2/3 (Grade 1: 0-4 PDCs; Grade 2: 5-9 PDCs; Grade 3: ≥10 PDCs) could effectively predict LNM. Martínez et al⁴¹ demonstrated that a combination of TB2/3 and high-grade PDC (positive PDC, consistent with the definition used in this study) could reduce unnecessary surgeries by 41.5% to 45% in pT1 CRC. Nevertheless, TB and PDC were still regarded as separate pathologic entities under standard grading criteria in these studies. In essence, these studies failed to capture the full morphologic spectrum of the tumor invasive front with one single pathologic parameter. To address these limitations, here we first defined PMCF as nonglandular malignant cellular foci consisting of ≥1 tumor cell at the tumor invasive front. Functioning as a unified biomarker, high-grade PMCF was found to be an independent risk factor with superior predictive performance. Beyond this, PMCF’s role as a holistic biomarker offers several practical advantages: it effectively eliminates the challenging and somewhat arbitrary distinction between 4-cell and 5-cell clusters, provides stronger LNM risk stratification than either TB or PDC with substantial reproducibility, captures the full morphologic spectrum and likely represents the biological continuum of the tumor invasive front. However, several practical problems should be considered: 1. Small malignant cellular foci that are indistinguishable from stromal components, such as macrophages, fibroblasts, inflammatory cells, or necrosis area should be discriminated to avoid overestimation of PMCF. 2. Malignant cellular foci within large mucin pools should not be classified as PMCF. 3. In case of numerous PMCF, poorly differentiated adenocarcinoma or micropapillary adenocarcinoma should be carefully distinguished.

The association between LNM risk factors and the number of metastatic lymph nodes remains unexplored. In our study, high-grade PMCF, absent BGA, positive CT examination, HER-2 with 2+/3+, Ki-67 ≥75%, and positive LVI demonstrated a stronger association with greater numbers of metastatic lymph nodes. Therefore, patients with these risk factors may require not only additional RS after ER but also more extensive lymph node dissection to reduce the risk of residual tumor.

To date, only the JSCCR has offered recommendations on the extent of lymph node dissection for pT1 CRC.² For clinically LNM-negative (cN(−)) patients, D2 lymph node dissection is recommended, whereas D3 dissection is recommended for cN(+) patients. Nonetheless, it remains unclear whether D3 dissection is truly necessary for pT1 CRC patients diagnosed as cN(+). Moreover, the patients who could benefit from D3 dissection have not been well defined. In our opinion, despite the extremely low rate of D3 LNM in pT1 CRC, routine D3 dissection is not applicable to all LNM-positive cases. Given that an appropriate extent of lymph node dissection can substantially reduce the risk of perioperative complications,^42,43 exploring risk factors for D3 LNM is of great significance. Our results revealed that high-grade PMCF, positive LVI, depressed tumor morphology, Ki-67 ≥75%, and absent BGA were significantly associated with an increased risk of D3 LNM. These findings suggest that patients with the above risk factors warrant close vigilance against potential D3 LNM. More rigorous evaluation for D3 LNM should be performed during both preoperative radiologic assessment and intraoperative examination to facilitate the extent of lymph node dissection.

Several limitations of this study should be acknowledged. First, this study prospectively enrolled a relatively modest number of patients from a single center. Thus, a larger prospective cohort from multicenter is required to validate the pathologic and clinical value of PMCF. Second, the event number with D3 metastasis might be insufficient to create a stable multivariate model for identifying the high-risk factors for D3 LNM. Thus, these findings must be interpreted cautiously. Expanding the sample size of D3 LNM patients is necessary in the future study. Third, this study included only Chinese cohorts. Future validation of the utility of PMCF in international cohorts is therefore warranted.

In conclusion, high-grade PMCF not only suggests additional RS after ER in pT1 CRC, but may also indicate D3 lymph node dissection. As a better pathologic predictor of LNM, PMCF provides an evidence-based rationale to optimize surgical decision-making in pT1 CRC.

Acknowledgments

The authors thank the staff from the department of pathology at all participating centers for providing pathologic slides.