This study was approved by the Ethics Committee of Xijing Hospital and informed consent was acquired for patient specimens and follow-up information. A total of 191 patients with CRC who had undergone surgical resection between June 2006 and November 2008 were investigated and followed up until 2016 in this study. According to the following inclusion and exclusion criteria (Figure 1), the postoperative specimens of 112 patients were involved. The inclusion criteria included: 1) histologically confirmed colorectal adenocarcinoma; 2) age ≥ 18; 3) without other malignant tumors. The exclusion criteria included: 1) non-adenocarcinoma (n = 15); 2) received preoperative anti-tumor therapies (n = 48); 3) follow-up data unavailable (n = 16); 4) unqualified specimens (AQP1, n = 9; AQP3, n = 5; AQP5, n = 3). Lastly, the total number of specimens evaluated for AQP1 expression was 103, AQP3 expression was 107 and AQP5 expression was 109. The main indicators were PFS and OS. PFS was defined as the period of time from the patient’s diagnosis date until any tumor progression, such as metastasis, recurrence, or death. OS was defined as the period of time from the diagnostic date until death from any cause.

Each TMA contained 17 patients’ tissues including 2 cores of tumor tissue and 1 core of paracancerous tissue for one patient. The core diameter of TMA was 1.6 mm. The immunohistochemical staining was performed on a Leica BOND-MAX autoimmunostainer (Leica Instrument Co., Ltd.). The tissue chips were first cut into 4-μm-thick slices using a microtome, attached to slides, dried, and baked at 60°C for 24 h before use. The prepared slices were dewaxed with xylene, rehydrated with different concentrations of ethanol (100% and 95%), rinsed with distilled water, and repaired with heat-induced antigen. Finally, the antigen was detected by an automatic staining machine (Benchmark) and Bond Polymer Purification Detection Kit (LeicaInstrumentCo., Ltd.). The primary antibodies (Abcam) were diluted as follows: anti-AQP1 (ab168387, 1:1,000), anti-AQP3 (ab125219, 1:500), and anti-AQP5 (ab92320, 1:200).

All slides were scanned with a Nano Zoomer SQ (HAMAMATSU, Japan). The expression levels of AQP1, 3 and 5 in the tumor region of each core of TMA were evaluated by Qupath software (Quantitative Pathology, version 0.2.0) and corresponding scores (Allred_score [21]and H_score [22]) were obtained. The average score of two tumor cores from the same patient was submitted for statistical analysis.

The combination of the values for staining intensity and the percentage of positively stained cells produced an Allred score that ranged from 0 to 8. The intensity was given a score between 0 and 3, with 0 representing no staining, 1 representing weak staining, 2 representing moderate staining, and 3 representing high staining Scores representing the proportion of positively stained tumor cells were stratified as 0 (no positive tumor cells), 1 (<1%), 2 (1%–10%), 3 (10%–33%), 4 (33%–67%) and 5 (67%–100%).

The H_score which ranged from 0 to 300 was calculated with the following equation: H_score = ∑ (PI  ×  I) = (percentage of cells with weak intensity  ×  1) + (percentage of cells with moderate intensity ×  2) + (percentage of cells with strong intensity × 3). The terms PI and I in the formula represent the proportion of positive cells to all cells and the staining intensity (0–3), respectively.

All analyses were completed using the statistical programming language R (version 4.1.2, 6.1, www.r-project.org) and SPSS (version 26.0, IBM, Armonk, NY). For all statistical tests, differences were considered statistically significant when the p-value was <0.05 (two-tailed).

The correlations between AQP expression and clinicopathological characteristics were investigated with chi-square test, t-test, or one-way ANOVA, when appropriate. The optimal cut-off values of AQP (Allred_score and H_score) to distinguish high AQP1, 3 and 5 expressions from low expression were determined separately using time-dependent ROC curves (R package “survival ROC”). The Kaplan-Meier curves and log-rank tests were used to perform survival analysis of AQP expression by R package “survival.” The Cox proportional hazard regression models were used to conduct both univariate and multivariate survival analyses.

A total of 112 patients were eventually enrolled in this retrospective study. The AQP1 group contained 103 consecutive patients (66 males, 37 females; mean age, 59.5 ± 13.4 years; mean follow-up, 78 months, range: 1–128 months). The AQP3 group contained 107 consecutive patients (67 males, 40 females; mean age, 58.8 ± 14.0 years; mean follow-up, 79 months, range: 1–128 months). The AQP5 group contained 109 consecutive patients (68 males, 41 females; mean age, 58.9 ± 14.0 years; mean follow-up, 78 months, range: 1–128 months). The other clinical characteristics of patients were presented in Table 1.

Patients were divided into high or low expression subgroups according to the optimal cut-off values (Allred_score and H_score) of AQP1, 3 and 5 expressions which were determined by time-dependent ROC curves according to the 5-year PFS. The optimal cut-off values for AQP1 (Allred_score) and AQP1 (H_score) were 6.0 and 116.17, respectively. The optimal cut-off values for AQP3 (Allred_score) and AQP3 (H_score) were 5.5 and 115.64, respectively. The optimal cut-off values for AQP5 (Allred_score) and AQP5 (H_score) were 6.0 and 18.64, respectively. Accordingly, Immunohistochemical examples of low and high expression of AQP1, 3 and 5 were shown in Figure 2.

This study investigated the relationship between AQP1, 3 and 5 expression and gender, age, primary tumor site, histological grading, and TNM stage. For Allred_score, the rate of AQP1 high expression in the positive lymph node group was significantly more common than that in the negative lymph node group (79.4% vs. 58.0%, χ² = 4.605, p = 0.032, Figure 3). For H_score, the expression of AQP1 was significantly higher in positive lymph node group than that in negative lymph node group (134.81 ± 44.573 vs. 105.978 ± 46.122, t = 3.016, p = 0.003, Figure 4).

The AQP3(Allred_score) was significantly correlated with the histological grading of CRC. (χ² = 10.773, p = 0.005, Supplementary Figure S1), AQP3 (H_score) had a similar conclusion (F = 4.212, p = 0.017, Supplementary Figure S2). In multiple comparisons between groups, AQP3 expression (H_score) was found to be significantly higher in well-differentiated group than in moderate-differentiated group (p = 0.005, Supplementary Figure S2), while no statistically significant differences in AQP3 expression were found between the well-differentiated group and poor-differentiated group, as well as moderate-differentiated group and poor-differentiated group (p = 0.233 and 0.542, respectively).

The AQP5 (Allred_score) was apparently associated with the tumor location of colorectal carcinoma (χ² = 8.123, p = 0.004, Supplementary Figure S3), AQP5 (H_score) had a similar conclusion, resulting in significantly lower AQP5 expression in the left hemicolectomy than in the right hemicolectomy (50.473 ± 54.175 vs. 90.052 ± 62.624, t = −3.531, p = 0.001, Supplementary Figure S4).

Kaplan-Meier survival curves demonstrated that the patient with high expression of the AQP1 had poorer 5-year PFS than patients with low expression of the AQP1 (Allred_score, 5-year PFS = 47% vs. 72%, p = 0.015, Figure 5A; H_score, 5-year PFS = 52% vs. 78%, p = 0.006; Figure 5B). Likewise, the survival rate of patients with high AQP1 expression in 5-year OS was significantly lower compared to that of patients with low AQP1 expression in 5-year OS (Allred_score, 5-year OS = 51% vs. 75%, p = 0.005, Figure 5C; H_score, 5-year OS = 56% vs. 80%, p = 0.002; Figure 5D). However, Kaplan-Meier survival curves presented that the 5-year PFS and OS of patients in high or low AQP3 and 5 expression groups had no statistical difference. (p > 0.05, Supplementary Figures S5, S6).

The AQP1, 3 and 5 expression and clinicopathologic factors, including age, gender, primary tumor site, histological grading, TNM-stage and treatment strategy, were submitted into univariate and multivariate COX regression analysis. The univariate Cox regression analysis showed that AQP1 (Allred_score and H_score), age, and TNM-stage were risk factors of 5-year PFS and OS for CRC (p < 0.05), while the expression of AQP 3 and AQP5 were not correlated with 5-year PFS and OS (p > 0.05, Table 2).

Since the Allred_score and H_score were two indicators to evaluate the same protein expression and the univariate COX analysis showed that the H_score was more significant than Allred_score in PFS and OS assessment (PFS, p = 0.008 vs. 0.017; OS, p = 0.003 vs. 0.008), the multivariate COX regression analysis was performed using H_score of AQP1 and indicated that AQP1 was no longer an independent risk factor for 5-year PFS (p = 0.086, HR = 1.866, HR95% CI: 0.916–3.800), but an independent risk factor for 5-year OS (p = 0.033, HR = 2.274, HR95% CI: 1.069–4.836, Table 3).