Post-operative anastomotic leakage and collagen changes in patients with rectal cancer undergoing neoadjuvant chemotherapy vs chemoradiotherapy

Abstract

Background

A significant difference in the anastomotic leakage (AL) rate has been observed between patients with locally advanced rectal cancer who have undergone preoperative chemotherapy and those undergoing preoperative chemoradiotherapy. This study aimed to quantitatively analyse collagen structural changes caused by preoperative chemoradiotherapy and illuminate the relationship between collagen changes and AL.

Methods

Anastomotic distal and proximal “doughnut” specimens from the Sixth Affiliated Hospital of Sun Yat-sen University (Guangzhou, China) were quantitatively assessed for collagen structural changes between patients with and without preoperative radiotherapy using multiphoton imaging. Then, patients treated with preoperative chemoradiotherapy were used as a training cohort to construct an AL–SVM classifier by the Mann–Whitney U test and support vector machine (SVM). An independent test cohort from the Fujian Province Cancer Hospital (Fuzhou, China) was used to validate the AL–SVM classifier.

Results

A total of 207 patients were included from the Sixth Affiliated Hospital of Sun Yat-sen University. The AL rate in the preoperative chemoradiotherapy group (n = 107) was significantly higher than that in the preoperative chemotherapy group (n = 100) (21.5% vs 7.0%, P = 0.003). A fully quantitative analysis showed notable morphological and spatial distribution feature changes in collagen in the preoperative chemoradiotherapy group. Then, the patients who received preoperative chemoradiotherapy were used as a training cohort to construct the AL–SVM classifier based on five collagen features and the tumor distance from the anus. The AL–SVM classifier showed satisfactory discrimination and calibration with areas under the curve of 0.907 and 0.856 in the training and test cohorts, respectively.

Conclusions

The collagen structure may be notably altered by preoperative radiotherapy. The AL–SVM classifier was useful for the individualized prediction of AL in rectal cancer patients undergoing preoperative chemoradiotherapy.

Introduction

Preoperative chemoradiotherapy can decrease tumor staging, shrink primary tumors, and increase the radical resection rate and anus preservation rate; hence, it has been recommended by the National Comprehensive Cancer Network (NCCN) guidelines as a standardized treatment strategy for locally advanced rectal cancer (LARC) [1, 2]. Anastomotic leakage (AL) is the most important complication after rectal cancer surgery and may increase the mortality and hospital stay of patients. Previous studies showed that there was no significant difference in outcomes between patients with LARC undergoing preoperative chemotherapy with and those without preoperative radiotherapy [1], and that preoperative radiotherapy increases the risk of AL after rectal cancer resection [3–5].

Collagen constitutes the majority of the extracellular matrix (ECM) and serves as a load-bearing “skeleton” in maintaining the mechanical stability of intestinal tissues [6]. In general, there is no difference in collagen levels in different parts of the human large intestine [7]. However, radiotherapy may cause excessive deposition and remodeling of collagen in the ECM due to tissue hypoxia effects [8], which may affect the mechanical stability of anastomosis. A recent study demonstrated that AL was associated with aggravated radiation damage [9]. However, there is currently a lack of a quantitative analysis to evaluate the specific changes in the collagen structure caused by radiotherapy and the relationship between the changes and AL.

Multiphoton imaging is a new imaging method based on nonlinear optical effects that combines two-photon excitation fluorescence (TPEF) and second harmonic generation (SHG) [10, 11]. Currently, multiphoton imaging has been widely applied in biological research [12–14]. In addition, SHG imaging can make the collagen structure visualized and provide multiple fully quantitative collagen features [15], including morphological and spatial distribution features, to diagnose several diseases and predict the outcomes [12–17].

Therefore, in this study, we used multiphoton imaging and full quantitative analysis to illuminate the specific collagen changes caused by radiotherapy and aimed to clarify the relationship between collagen changes and AL in rectal cancer patients who underwent preoperative chemoradiotherapy.

Methods

Patients and specimens

This study was approved by the Institutional Review Board at the Sixth Affiliated Hospital of Sun Yat-sen University, Guangzhou, China (Center 1) and Fujian Province Cancer Hospital, Fuzhou, China (Center 2) (K2022-056–01). The inclusion criteria were (i) aged >18 and <75 years; (ii) diagnosed with rectal cancer by pathology; (iii) diagnosed with stage II or stage III by imaging, including computed tomography (CT), magnetic resonance imaging (MRI), and endoscopic ultrasonography; (iv) tumor distance from the anus ≤12 cm; and (v) underwent radical anterior rectal resection after neoadjuvant treatment. The exclusion criteria were (i) Miles or Hartmann operation; (ii) short-term preoperative radiotherapy; or (iii) positive margin. Multiphoton imaging was performed on the distal and proximal “doughnut” specimens. Patient informed consent was waived by the institutional review board. The study was conducted in compliance with the Declaration of Helsinki. Each specimen underwent routine pathological procedures, including 10% buffered formalin treatment, paraffin embedding, and 5-µm sections for imaging, and then the multiphoton images were compared using hematoxylin-eosin staining (HE) images for histologic assessment.

We collected baseline clinicopathologic data of each patient, including age, sex, body mass index, preoperative albumin (ALB) level, preoperative hemoglobin (HGB) level, American Society of Anesthesiologists score, tumor distance from the anus, diverting stoma, tumor size, TNM stage (American Joint Committee on Cancer, 8th edition), and pathological complete response (pCR) from the institutional database of the two centers.

Treatment

The preoperative chemotherapy regimen (mFOLFOX 6) was scheduled as a 120-minute oxaliplatin infusion at a dose of 85 mg/m² with concurrent intravenous infusion of leucovorin 400 mg/m² followed by 5-fluorocrail given every 2 weeks. The 5-fluorocrail was given as a 400-mg/m² intravenous bolus injection followed by a 46-hour continuous intravenous infusion at a dose of 2,400 mg/m². Drugs were given for five cycles in the preoperative chemoradiotherapy group and for four to six cycles in the preoperative chemotherapy group. All patients in the preoperative chemoradiotherapy group received the same radiotherapy regimen, which consisted of intensity modulation radiotherapy at a total dose of 50 Gy, five times a week for 5 weeks, starting at the second chemotherapy cycle. The clinical targets included mesenteric and pelvic lymphatic areas, and all radiation oncologists received standard training [1]. The patients underwent radical surgery 6–8 weeks after finishing chemoradiotherapy or 4 weeks after finishing chemotherapy.

In the present study, all procedures were strictly performed according to the principle of total mesorectal excision (TME) by senior attending surgeons specializing in colorectal surgery, who had performed >1,000 laparoscopic rectal cancer surgeries. Moreover, standardized post-operative management and monitoring were performed for the patients after surgery. The adjuvant chemotherapy regimen started within 6 weeks after surgery, and the regimen was the same as the preoperative chemotherapy regimen.

The definition of AL was proposed in 2010 by the International Study Group of Rectal Cancer [18]: (i) the anastomosis site was defective in the intestinal wall, resulting in communication with the intestine; and (ii) pelvic abscesses near the anastomosis were also classified as AL. All cases of AL were diagnosed by one or more of the following methods: digital rectal examination, endoscopy, imaging (CT or MRI), or operative finding. Patients with Grades A–C AL were all included in this study.

Multiphoton image acquisition and collagen feature extraction

Panoramic SHG imaging was performed on all specimens. Since collagen is concentrated in the submucosa and provides structural support for the large intestine tissue [6], the collagen structure in the submucosa was analysed in this study. An independent pathologist who was blinded to the clinical characteristics and outcomes reassessed the region of the submucosa in the HE image. Then, an optical expert selected the collagen in the submucosa of the SHG image for feature extraction according to the HE image. Our previous study demonstrated that multiphoton images are very sensitive to collagen and are comparable to HE images [13]. The flow chart of this study is shown in Figure 1. Multiphoton imaging can objectively reflect the collagen structure of the doughnut. Collagen quantification and subsequent analysis were performed based on multiphoton images, so the whole study was performed objectively.

Figure 1.

Flow chart of this study. The images in the left panel are the HE image and the corresponding multiphoton image, and the SHG image (collagen is presented in green) of the doughnut specimen. Multiphoton images can objectively reflect the collagen structure of the doughnut. Subsequent analysis was performed based on collagen quantification of the multiphoton image. Then, 142 collagen features of the distal and proximal doughnuts were compared between patients in the preoperative chemoradiotherapy group and the preoperative chemotherapy group. Finally, the Mann–Whitney U test and SVM–RFE were applied to select predictive features to construct the AL–SVM classifier for predicting AL in patients with preoperative chemoradiotherapy and was validated in the test cohort. Scale bars: 1,000 μm. HE, hematoxylin-eosin; SHG, harmonic generation; SVM–RFE, support vector machine–recursive feature elimination; AL–SVM, anastomotic leakage–support vector machine.

The nonlinear optical microscope is constructed using a commercial laser scanning microscope (LSM 880; Zeiss, Jena, Germany) and a mode-locked titanium:sapphire femtosecond laser (140 fs, 80 MHz) and can be tuned in the range of 680–1,080 nm [19]. The excitation laser (800 nm) was focused on the tissue through a flat apochromatic objective (10×, numerical aperture [NA] = 0.8). The two-channel mode could achieve TPEF and SHG, in which one channel corresponds to a wavelength range of 430–708 nm to show the morphologies of tissue components from the TPEF signals (red), whereas another channel covering the wavelength range from 387 to 409 nm presents the microstructures of collagen components from the SHG signals (green).

MATLAB 2015b (MathWorks) was used to extract collagen features as previously described [17]. A total of 142 collagen features were extracted from the SHG images used to describe the spatial distribution and morphological features of the collagen (Supplementary Table 1). Eight morphological features were the collagen area, number, length, width, straightness, cross-link density, cross-link space, and orientation. Six histogram features were the mean, variation, skewness, kurtosis, energy, and entropy. We also included an 80 gray-level co-occurrence matrix (GLCM). The contrast, correlation, energy, and homogeneity were calculated from the GLCM with five different displacements of pixels at 1, 2, 3, 4, and 5, and four different directions at 0, 45, 90, and 135 degrees [20]. To calculate the 48 Gabor wavelet transform features, SHG images were convolved with Gabor filters at four different scales and six different orientations, and the mean and variation of the magnitude of the convolution over the image at each setting were calculated.

Support vector machine

To reduce overfitting or any type of bias in our prediction model, two steps were used for collagen feature selection. First, patients who had received neoadjuvant radiotherapy from Center 1 were used as a training cohort. A univariate statistical test (Mann–Whitney U test) was used to select potential collagen predictors from 284 collagen features (142 from the distal doughnut and 142 from the proximal doughnut) between the AL and non-AL groups. Second, the support vector machine–recursive feature elimination (SVM–RFE) algorithm was used to select and rank potential collagen and clinicopathological predictors to construct an AL–SVM classifier in the training cohort [21–23], and the bootstrap method was used for internal validation [16]. Then, the AL–SVM classifier was confirmed in the test cohort from Center 2.

In the AL–SVM classifier, patients with a high probability of AL on one side of the SVM hyperplane were defined as the high AL–SVM classifier and the other side was defined as the low AL–SVM classifier. The area under the receiver-operating characteristic (ROC) curve was used to assess the discrimination of the AL–SVM classifier. A calibration curve was used to assess its calibration.

In addition, decision curve analysis (DCA) was used to quantitatively analyse the net benefit of different threshold probabilities to evaluate the clinical usefulness of the AL–SVM classifier [24, 25].

Statistical analysis

All statistical analyses were performed using SPSS version 24.0 software (IBM, Armonk, New York, USA) and R software (version 4.0.2; http://www.Rproject.org). The chi-square test and Fisher's exact test were used to analyse the categorical variables. The Mann–Whitney U test was performed to analyse the non-normally distributed parameters, and Student’s t-tests were used to analyse the normally distributed parameters. Univariate and multivariate logistic regression with backward stepwise elimination were used to select the predictors and calculate the odds risk (OR) with the 95% confidence interval (CI). All tests were two-tailed, and a P-value of <0.050 was determined to be statistically significant. The SVM was analysed using the “e1071” package. The calibration curve was performed using the “rms” package, the ROC curve was developed using the “pROC” package, and DCA was performed using the “rmda” package.

Results

Patient demographics and cancer characteristics

Based on the inclusion and exclusion criteria, 207 patients were involved in this study between 1 January 2018 and 31 August 2020. Among them, 107 patients received preoperative chemoradiotherapy and 100 received preoperative chemotherapy from Center 1 (Supplementary Figure 1). The baseline characteristics of the 207 patients are shown in Table 1. The proportion of AL patients in the preoperative chemoradiotherapy group was significantly higher than that in the preoperative chemotherapy group (21.5% vs 7%, P = 0.003). Univariate analysis showed that preoperative radiotherapy (P = 0.005) and tumor distance from the anus (P = 0.002) were significantly associated with AL in all patients. Multivariable analysis demonstrated that preoperative radiotherapy (OR, 3.794; 95% CI, 1.514–9.505; P = 0.004) and tumor distance from the anus of <5 cm (OR, 3.768; 95% CI, 1.661–8.547; P = 0.001) were independent risk factors for AL (Supplementary Table 2).

Table 1.

Baseline characteristics of 207 patients from the Sixth Affiliated Hospital of Sun Yat-sen University, Guangzhou, China (Center 1)

Characteristic	Preoperative chemoradiotherapy	Preoperative chemotherapy	P
	(n = 107)	(n = 100)
Age (years)			0.103
≥65	17 (15.9)	25 (25.0)
<65	90 (84.1)	75 (75.0)
Sex			0.728
Male	72 (67.3)	65 (65.0)
Female	35 (32.7)	35 (35.0)
BMI			0.070
≥24	34 (31.8)	44 (44.0)
<24	73 (68.2)	56 (56.0)
ALB (g/L)			0.597
≥35	95 (88.8)	90 (90.0)
<35	12 (11.2)	10 (10.0)
HGB (g/L)			0.924
≥120	56 (52.3)	53 (53.0)
<120	51 (47.7)	47 (47.0)
ASA score			0.713
1	16 (15.0)	17 (17.0)
2	87 (81.3)	81 (81.0)
3	4 (3.7)	2 (2.0)
Tumor distance from the anus (cm)			0.783
≥5	73 (68.2)	70 (70.0)
<5	34 (31.8)	30 (30.0)
Diverting stoma			0.001
Yes	92 (86.0)	64 (64.0)
No	15 (14.0)	36 (36.0)
Tumor size (cm)			0.006
≥3	22 (22.6)	38 (38.0)
<3	85 (79.4)	62 (62.0)
ypT			0.013
0–II	57 (53.3)	36 (36.0)
III–IV	50 (46.7)	64 (64.0)
ypN			0.006
N0	62 (57.9)	39 (39.0)
N0+	45 (42.1)	61 (61.0)
ypM			0.143a
M0	107 (100)	98 (98.0)
M1	0 (0)	2 (2.0)
AJCC stage			0.004
0–II	62 (57.9)	38 (38.0)
III–IV	45 (42.1)	62 (62.0)
pCR			0.001
Yes	30 (28.0)	10 (10.0)
No	77 (72.0)	90 (90.0)
AL			0.003
Yes	23 (21.5)	7 (7.0)
No	84 (78.5)	93 (93.0)

Data are presented as number of patients followed by percentage (%).

BMI, body mass index; ALB, albumin; HGB, hemoglobin; ASA; American Society of Anesthesiologists; AJCC, American Joint Committee on Cancer (8th edition); pCR, pathological complete response; AL, anastomotic leakage.

^aData were calculated using Fisher’s exact test.

Differences in the collagen structure between patients in the preoperative chemoradiotherapy group and the preoperative chemotherapy group

Figure 2 shows two patients' representative images. Both patients were 50 years old with mid-rectal cancer. One received preoperative chemoradiotherapy and the other received preoperative chemotherapy. No significant collagen differences were observed in the HE images; however, the SHG images clearly showed collagen differences between the two patients.

Figure 2.

Representative images of two 50-year-old male patients with mid-rectal cancer. Representative HE image, multiphoton image, SHG image, and binary image in the distal (A)–(D) and proximal (E)–(H) doughnut of the same patient with preoperative chemoradiotherapy. Representative HE image, multiphoton image, SHG image, and binary image in the distal (I)–(L) and proximal (M)–(P) doughnut of the same patient with preoperative chemotherapy. Collagen is represented as green in the multiphoton and SHG images and white in the binary image. HE images have difficulty distinguishing the collagen structure, but multiphoton images and SHG images can clearly show the structural features of collagen without the need for fluorescence agents. Scale bars: 1,000 μm. HE, hematoxylin-eosin; SHG, harmonic generation. (A color version of this figure appears in the online version of this article)

Collagen features were assessed in the distal and proximal doughnuts in patients with or without preoperative radiotherapy. The results showed that collagen morphological and spatial features were significantly different between the two groups in both the distal and proximal doughnuts (Supplementary Tables 3–6). These results suggest that notable morphological and spatial distribution feature changes in collagen may be induced by preoperative radiotherapy.

AL–SVM classifier to predict AL in patients with preoperative chemoradiotherapy

Patients with preoperative chemoradiotherapy were used as the training cohort and divided into AL and non-AL subgroups to assess the relationship between collagen structural features and the occurrence of post-operative AL. We found that 19 collagen features were significantly different between the AL and non-AL subgroups of patients with preoperative chemoradiotherapy (P < 0.05). Among the 19 features, 13 were from the distal doughnut and 6 were from the proximal doughnut (Table 2). Then, collagen features were divided into high- and low-feature groups based on the cut-off value of selected collagen features. Next, SVM–RFE was used to select predictors from potential collagen and clinicopathological predictors (Supplementary Figure 2). As a result, tumor distance from the anus, collagen area, number of collagen fibers, collagen cross-link density, collagen orientation, and GLCM 25 from the distal doughnut were used to construct the AL–SVM classifier. The associations between the selected predictors and AL are shown in Figure 3.

Table 2.

Relationship between the collagen structural features of the doughnut and AL in patients with preoperative chemoradiotherapy in the training cohort

Feature	AL	Non-AL	P
	(n = 23)	(n = 84)
Distal doughnut
Area	0.220 (0.174, 0.311)	0.165 (0.114, 0.220)	0.007
Number	0.017 (0.013, 0.021)	0.011 (0.009, 0.016)	0.007
Length	44.569 (39.640, 45.820)	48.355 (41.973, 54.342)	0.042
Straightness	0.949 (0.942, 0.954)	0.941 (0.932, 0.950)	0.003
Cross-link density	0.039 (0.036, 0.041)	0.037 (0.035, 0.039)	0.025
Orientation	0.826 (0.810, 0.857)	0.803 (0.749, 0.831)	0.014
GLCM25	0.023 (0.017, 0.031)	0.018 (0.015, 0.029)	0.020
GLCM42	0.477 (0.439, 0.488)	0.456 (0.435, 0.473)	0.046
GLCM61	0.044 (0.035, 0.064)	0.037 (0.029, 0.061)	0.021
GLCM73	0.040 (0.031, 0.059)	0.034 (0.027, 0.056)	0.049
Gabor8	0.207 (0.177, 0.224)	0.185 (0.176, 0.208)	0.038
Gabor33	0.136 (0.114, 0.150)	0.123 (0.116, 0.133)	0.050
Gabor36	0.369 (0.324, 0.425)	0.329 (0.314, 0.391)	0.034
Proximal doughnut
Skewness	16.998 (16.202, 21.532)	15.712(14.359, 20.042)	0.034
GLCM25	0.018 (0.013, 0.031)	0.013 (0.011, 0.025)	0.038
GLCM42	0.451 (0.441, 0.477)	0.445(0.429, 0.470)	0.048
GLCM65	0.028 (0.021, 0.060)	0.020 (0.017, 0.047)	0.032
Gabor8	0.202 (0.171, 0.228)	0.179 (0.166, 0.203)	0.041
Gabor33	0.126 (0.102, 0.142)	0.115 (0.103, 0.127)	0.042

All values are presented as median followed by range.

AL, anastomotic leakage; GLCM; gray-level co-occurrence matrix.

Figure 3.

Correlations among AL and the AL–SVM classifier and predictors. (A) Distribution of the AL–SVM classifier, six predictors, and AL of each patient. Different variables are represented by different colors. (B) Box plots showing group differences of five selected collagen features between the AL vs non-AL groups. Box Whisker plots with median (2.5th, 25th, 75th, and 97.5th percentiles) and outliers. AL–SVM, anastomotic leakage–support vector machine.

Furthermore, the AL–SVM classifier was validated in a test cohort from Center 2 that contained 81 consecutive patients undergoing preoperative chemoradiotherapy. The incidence of AL in the test cohort was 17.3% (14/81), which was not significantly different from that in the training cohort (P = 0.472). In addition, the patient demographics and cancer characteristics were also similar between the two cohorts (Supplementary Table 7).

The AL–SVM classifier displayed satisfactory discrimination and calibration. The area under the curve (AUC) of the AL–SVM classifier was 0.907 (95% CI, 0.841–1.000) in the training cohort (Figure 4A). Compared with the other individual characteristics, the AL–SVM classifier had significantly improved the power of discrimination (Table 3). The calibration curve demonstrated that the predicted AL probability was close to the actual AL probability (Figure 4B). The bootstrap method was performed for 1,000 resamples used for internal verification and the results showed that the average consistency index was 0.897. These results were also presented in the test cohort (Figure 4C and D) and the AUC of the AL–SVM classifier was 0.856 (95% CI, 0.727–0.988) (Table 3). The DCA showed that using the AL–SVM classifier to predict AL in patients with preoperative chemoradiotherapy could add more benefit than either the treat-all-patients scheme or the treat-none scheme in the two cohorts (Figure 4E and F).

Figure 4.

Performance of the AL–SVM classifier. The ROCs of the AL–SVM classifier, five collagen structural features, and tumor distance from the anus in the training (A) and test cohorts (B). Different variables are represented by different colors. The calibration curves of the AL–SVM classifier in the training (C) and test cohorts (D) indicate the consistency of the predicted and observed results. The y-axis represents the actual AL probability, the x-axis represents the predicted AL probability, and the black diagonal dotted line represents the perfect prediction model. The solid red line represents the performance of the AL–SVM classifier; a closer fit to the diagonal black dotted line indicates that the AL–SVM classifier has good calibration. In the DCA of the AL–SVM classifier in the training (E) and test cohorts (F), the y-axis indicates the net benefit. The red line indicates the AL–SVM classifier. The green line represents the assumption that all patients had AL. The black line represents the assumption that no patients had AL. These DCAs showed that using the AL–SVM classifier to predict AL in patients with preoperative radiotherapy can add more benefit than either the treat-all-patients scheme or the treat-none scheme in both cohorts. AL–SVM, anastomotic leakage–support vector machine; ROC, receiver-operating characteristic curve; DCA, decision curve analysis.

Table 3.

ROC curves for the AL–SVM classifier and other characteristics as predictors of AL in the training cohort and the test cohort

Characteristic	Training cohort	P	Test cohort	P
	AUC (95% CI)		AUC (95% CI)
AL–SVM classifier	0.907 (0.814–1.000)	Ref.	0.856 (0.727–0.988)	Ref.
Collagen area	0.683 (0.557–0.810)	<0.001	0.678 (0.525–0.831)	<0.001
Number of collagen	0.683 (0.565–0.801)	<0.001	0.669 (0.521–0.817)	<0.001
Collagen cross-link density	0.662 (0.526–0.797)	<0.001	0.674 (0.509–0.839)	<0.001
Collagen orientation	0.671 (0.552–0.790)	<0.001	0.684 (0.537–0.830)	<0.001
GLCM 25	0.655 (0.539–0.771)	<0.001	0.633 (0.477–0.790)	<0.001
Tumor distance from the anus	0.630 (0.496–0.763)	<0.001	0.650 (0.490–0.810)	<0.001

ROC, receiver-operating characteristic curve; AL–SVM, anastomotic leakage–support vector machine; AUC, area under the curve; CI, confidence interval; Ref., reference; GLCM, gray-level co-occurrence matrix.

In addition, the AL–SVM classifier showed satisfactory sensitivity, specificity, accuracy, positive predictive value, and negative predictive value in the training cohort, test cohort, and all patients (Table 4).

Table 4.

The performance of the AL–SVM classifier in estimating the risk of AL

Variable	Training cohort	Test cohort	All patients
Sensitivity, %	82.6 (62.9–93.0)	78.6 (52.4–92.4)	81.1 (65.8–90.5)
Specificity, %	98.8 (93.6–99.9)	92.5 (83.7–96.8)	96.0 (91.3–98.2)
Accuracy, %	95.3 (89.5–98.0)	90.1 (81.7–94.9)	93.1 (88.5–95.9)
PPV, %	95.0 (76.4–99.7)	68.8 (44.4–85.8)	83.3 (68.1–92.1)
NPV, %	95.4 (88.8–98.2)	95.4 (87.3–98.7)	95.3 (90.6–97.8)

AL–SVM, anastomotic leakage–support vector machine; PPV, positive predictive value; NPV, negative predictive value.

Discussion

Preoperative chemoradiotherapy increases the occurrence rate of post-operative AL. To clearly illuminate specific changes in the collagen structure caused by preoperative chemoradiotherapy, multiphoton imaging was used to perform imaging. A full quantitative analysis of collagen in the distal and proximal doughnuts of patients with or without preoperative radiotherapy was conducted. Our results suggested that notable morphological and spatial distribution of collagen changes can be caused by preoperative radiotherapy. Then, we developed an AL–SVM classifier and validated its satisfactory discrimination and calibration. The AL–SVM classifier integrates predictive collagen features and clinicopathological characteristics and may be used to individually predict the risk of AL in LARC patients treated with preoperative chemoradiotherapy.

Collagen, the main component of the ECM, plays a vital role in maintaining the integrity and structure of intestinal tissue and is one of the most important components of the mechanical stability of intestinal anastomosis [7, 26–28]. In general, there are no differences in collagen levels between different segments of the healthy human colon [7]. However, radiotherapy could cause changes in the structure of collagen, including remodeling, deposition, breakage, and increased cross-linking [8, 29].

Multiphoton imaging provides a powerful tool to evaluate collagen structure in patients treated with or without preoperative radiotherapy. In this work, the high-dimensional features of collagen, including morphological and spatial distribution features from multiphoton images, could be extracted in the doughnut. This is conducive to more accurate and objective statistical analysis.

Previous studies have found that preoperative chemotherapy does not increase the occurrence rate of AL compared with surgery alone [30, 31]. Qin et al. [9] constructed a specific scoring system to assess the histopathological features in patients treated with preoperative chemoradiotherapy, preoperative chemotherapy, and surgery alone. The results showed that the histopathological scores for the distal and proximal margins of patients treated with preoperative chemoradiotherapy were significantly higher than those of patients treated with preoperative chemotherapy and those of patients treated with surgery alone, whereas there was no significant difference between the patients treated with preoperative chemotherapy and the patients treated with surgery alone. Therefore, the difference in histopathological features was caused by radiotherapy [9]. Since previous studies have already shown that there is no significant difference in the occurrence rate of AL between the preoperative chemotherapy group and the surgery alone group, our current research was focused on the difference between the preoperative chemoradiotherapy group and the preoperative chemotherapy group. In this study, one group was treated with preoperative chemoradiotherapy and another group was treated with preoperative chemotherapy alone. Both groups received the same chemotherapy regimen; the preoperative chemoradiotherapy further received radiotherapy. We directly compared the collagen structure of the distal and proximal doughnuts between the patients treated with or without preoperative radiotherapy and found that the collagen structure showed notable morphological and spatial distributions of collagen in both the distal and proximal doughnuts. Morphologically, the collagen structure of the doughnut in patients undergoing radiotherapy had more collagen content; had more collagen fibers; had shorter, wider, greater cross-linking density; was more curved; and was more disordered than that in patients not undergoing radiotherapy. Moreover, histogram features, GLCM texture features, and Gabor wavelet transform features were used to describe the spatial distribution of collagen from different perspectives. Similarly, the spatial distribution of collagen was significantly changed by radiotherapy. The support and physiological function of collagen are affected by notable structural changes induced by radiotherapy [32]. These results demonstrated that the changes in collagen structure were caused by radiotherapy and may potentially cause the increased occurrence rate of preoperative AL.

Then, we further compared the collagen features between AL and non-AL patients treated with preoperative chemoradiotherapy. We obtained a total of 284 collagen features from each patient, including 142 from the distal doughnut and 142 from the proximal doughnut. The Mann–Whitney U test showed that 19 collagen features were significantly different between AL and non-AL patients. Then, the AL–SVM classifier was developed by the SVM–RFE algorithm. The performance of the AL–SVM classifier showed satisfactory discrimination and calibration in the training and test cohorts. The DCA determined that the AL–SVM classifier has clinical application value [24, 33]. This will be useful for better selection and management of patients who receive preoperative chemoradiotherapy.

Previous studies indicated that the predictive ability of a single parameter was insufficient [33–35]. Compared with other machine-learning algorithms, SVM is more suitable for managing classification based on high-dimensional data and limited training samples to select the most valuable predictors. The AL–SVM classifier integrates the tumor distance from the anus and five collagen features from the distal doughnut. Moreover, the AL–SVM classifier had a stronger discriminative ability than any single parameter. These results showed that patients who received preoperative radiotherapy with a lower tumor location and higher collagen area, a greater proportion of collagen fibers, higher collagen cross-link density, disorder collagen orientation, and higher GLCM 25 were classified into the high AL–SVM group and were highly likely to develop AL after surgery. This result indicated that changes in collagen caused by preoperative radiotherapy at the distal doughnuts were associated with post-operative AL. This may be related to the choice of margin. In the clinic, surgeons usually use part of the rectosigmoid or sigmoid colon 10–15 cm away from the edge of the tumor as the proximal margin. However, the distal margin was only 1–2 cm away from the edge of the tumor according to NCCN guidelines [36] and the distal rectum was exposed to radiation and thus was unavoidably damaged. Hence, the collagen structure at the distal doughnut was affected by radiotherapy, which was associated with post-operative AL. Intraoperative freezing of the distal doughnut assessed by multiphoton imaging could be used to evaluate the extent and severity of radiotherapy damage. If severely damaged doughnuts tend to have a higher risk of AL, more distal margins should be removed during surgery. For these patients, the surgeon should release more rectal tissue and perform the anastomosis at a more distant location to assiduously avoid anastomosing the colon to the radiated-injured rectum, which may be an effective method of reducing the occurrence rate of AL. The major clinical crux of this research study is related to both radiation-damaged tissue and the downstream effect of AL. Our findings suggest that using multiphoton imaging to determine the extent of radiation damage and to determine distal margins will be very useful for tailored surgery.

Despite the encouraging results, our current study is limited by its retrospective nature. First, all clinicopathological characteristics and specimens were retrospectively obtained from two Chinese medical institutions in this study. Therefore, potential bias was inevitable. Second, the distribution of clinicopathological characteristics might be different in other areas. Thus, a prospective, international, multicenter clinical trial will be needed to further verify our findings. Third, radical surgery should be performed 5–12 weeks following preoperative chemoradiotherapy according to NCCN guidelines [37]. Although an increased interval time may improve the pCR rate [38], it can also cause more severe tissue fibrosis in the radiotherapy area, thus leading to poorly delineated anatomical gaps, increasing the difficulty of surgery and the risk of post-operative complications, and decreasing the R0 resection rate and the quality of TME surgery [39, 40]. In this retrospective study, radical surgery was performed 6–8 weeks after chemoradiotherapy. We did not compare collagen changes between 6–8 and 12–20 weeks after chemoradiotherapy, which needs to be further studied.

In conclusion, our study suggests that collagen structure in the ECM could be notably altered due to preoperative radiotherapy. The AL–SVM classifier, based on the collagen structure and tumor distance from the anus, may be useful for individualized prediction of AL in LARC patients with preoperative chemoradiotherapy.

Supplementary data

Supplementary data is available at Gastroenterology Report online.

Authors’ Contributions

W.J., H.M.W., J.X.Z., S.M.Z., H.W., and J.Y. conceived and designed the project. W.J., H.M.W., J.X.Z., Y.D.Z., and S.Y.X. collected the data. W.J., H.M.W., J.X.Z., and S.Y.X. analysed and interpreted the data. W.J., H.M.W., and J.X.Z. drafted the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by grants from the National Natural Science Foundation of China [No. 82273360], the State’s Key Project of Research and Development Plan [No. 2019YFE0113700], the Guangdong Provincial Key Laboratory of Precision Medicine for Gastrointestinal Cancer [No. 2020B121201004], the Guangdong Provincial Major Talents Project [No. 2019JC05Y361], the China Postdoctoral Science Foundation [No. 2020M682789], the Natural Science Foundation of Fujian Province [No. 2018J07004], the Joint Funds of Fujian Provincial Health and Education Research [No. 2019-WJ-21], the Science and Technology Program of Fujian Province [No. 2018Y2003, 2019L3018, and 2019YZ016006], the Clinical Research Startup Program of Southern Medical University by High-level University Construction Funding of Guangdong Provincial Department of Education [No. LC2016PY010], the Clinical Research Project of Nanfang Hospital [No. 2018CR034, 2020CR001, and 2020CR011], the President Foundation of Nanfang Hospital, Southern Medical University [No. 2019Z023], and the Training Program for Undergraduate Innovation and Entrepreneurship [No. 201912121008, 202012121091, and 202012121277].

Conflict of Interest

None declared.