Overcoming chemoresistance in non-angiogenic colorectal cancer by metformin via inhibiting endothelial apoptosis and vascular immaturity

Abstract

The development of chemoresistance which results in a poor prognosis often renders current treatments for colorectal cancer (CRC). In this study, we identified reduced microvessel density (MVD) and vascular immaturity resulting from endothelial apoptosis as therapeutic targets for overcoming chemoresistance. We focused on the effect of metformin on MVD, vascular maturity, and endothelial apoptosis of CRCs with a non-angiogenic phenotype, and further investigated its effect in overcoming chemoresistance. In situ transplanted cancer models were established to compare MVD, endothelial apoptosis and vascular maturity, and function in tumors from metformin- and vehicle-treated mice. An in vitro co-culture system was used to observe the effects of metformin on tumor cell-induced endothelial apoptosis. Transcriptome sequencing was performed for genetic screening. Non-angiogenic CRC developed independently of angiogenesis and was characterized by vascular leakage, immaturity, reduced MVD, and non-hypoxia. This phenomenon had also been observed in human CRC. Furthermore, non-angiogenic CRCs showed a worse response to chemotherapeutic drugs in vivo than in vitro. By suppressing endothelial apoptosis, metformin sensitized non-angiogenic CRCs to chemo-drugs via elevation of MVD and improvement of vascular maturity. Further results showed that endothelial apoptosis was induced by tumor cells via activation of caspase signaling, which was abrogated by metformin administration. These findings provide pre-clinical evidence for the involvement of endothelial apoptosis and subsequent vascular immaturity in the chemoresistance of non-angiogenic CRC. By suppressing endothelial apoptosis, metformin restores vascular maturity and function and sensitizes CRC to chemotherapeutic drugs via a vascular mechanism.

Graphical abstract

Highlights

• •CRCs with a non-angiogenic vascular phenotype respond poorly to chemotherapy in vivo.
• •Tumor cells induce endothelial apoptosis and vascular immaturity, thus limiting chemo-drug delivery.
• •Metformin sensitizes non-angiogenic CRCs to chemo-drug by suppressing endothelial apoptosis via elevating MVD and improving vascular maturity.
• •Endothelial apoptosis was mediated by the activation of caspase signaling, which was abrogated by metformin.

1. Introduction

Colorectal cancer (CRC) is a common gastrointestinal malignancy with a high mortality rate [[1], [2], [3]], which seriously affects patient prognosis and quality of life. Over the past decades, chemotherapy has become one of the standard treatments for CRC [4]. Although early screening techniques have improved the 5-year survival rate of patients with CRC, many patients are diagnosed at advanced stages [5]. Indeed, many CRC patients have a lower survival rate because they do not respond to chemotherapy, which is mediated by chemoresistance [4]. Unfortunately, CRC patients also have a low response rate to novel immune checkpoint therapy [6]. Thus, it is of paramount importance to elucidate the mechanism of chemoresistance in patients with CRC.

Chemoresistance, which includes inherent and acquired drug resistance, has been demonstrated to be regulated by multiple factors, such as molecular targets, drug uptake, drug activation, and drug metabolism [7]. In addition, abnormalities in the tumor vasculature contribute to chemoresistance by affecting vasculature-mediated drug delivery [8]. According to the activation status of endothelial cells (ECs), tumor growth could be independent of or dependent on angiogenesis (angiogenic or non-angiogenic phenotype) [9,10]. It has been reported that metastatic CRCs with a non-angiogenic histological growth pattern (HGP) respond poorly to chemotherapy, compared with those with angiogenic HGP [11]. Unlike angiogenic HGP, the growth of non-angiogenic metastatic CRC depends on the co-option of pre-existing vessels in peritumoral tissues [12]. However, whether or how this vascular phenotype affects chemoresistance in CRC remains unclear.

Once a chemo-drug enters the body, it must be delivered into the tumor to mediate the cytotoxic effect by the vessel [13,14]. Although co-opted vessels have vascular maturity similar to that of normal tissues, non-angiogenic cancers are not more responsive to chemotherapy than angiogenic cancers [9,15]. Strilic et al. [16] reported that tumor cells were capable of inducing cell death in ECs, which affected the structure of the tumor vasculature and reduced microvessel density (MVD) in a non-angiogenic model. However, it remains unclear whether EC death-induced MVD decreases cancer chemosensitivity. In addition, very few studies have focused on whether the vascular structure and maturity and status of the component cells are affected after the vessels are co-opted by tumor cells.

Metformin, a first-line antidiabetic agent, has been demonstrated to protect ECs against endothelial apoptosis by phosphorylating adenosine monophosphate-activated protein kinase (AMPK) [14,[17], [18], [19]]. In the current paper, our results showed that non-angiogenic CRCs were resistant to chemotherapy, as MVD was reduced by endothelial apoptosis. By suppressing tumor cell-induced endothelial apoptosis, metformin ameliorates chemoresistance by increasing MVD and restoring vascular function and maturity. These findings may contribute to the understanding of non-angiogenic CRC following conventional chemotherapy.

2. Methods and materials

2.1. Chemicals and reagents

Drabkin's reagent kit 525 for hemoglobin determination was obtained from Sigma-Aldrich (Sigma-Aldrich (Shanghai) Trading Co., Shanghai, China). Cyclophosphamide (CTX, No. 13849), metformin (No. 13118), and 5-fluoro-pyrimidinedione (5-FU, No. 14416) were purchased from Cayman Chemical Company (Ann Arbor, MI, USA). The preactivated forms of CTX were ozonized to generate activated 4-OOH-CTX for in vitro experiments as previously described [20,21].

2.2. Cells culture and proliferation assays

Human umbilical vascular endothelial cells (HUVECs), murine 4T1 metastatic breast cancer, murine CT-26 CRC, as well as murine B16 melanoma cancer and human SW-480 and SW-620 CRC cell lines were obtained from the American Type Culture Collection (ATCC, Manassas, VA, USA). Short tandem repeat genotyping analysis and Mycotest were performed to detect cell line unintended cross-contamination and mycoplasma contamination before experiments, respectively. For proliferation assay, all cell lines were cultured in Dulbecco's Modified Eagle's medium (DMEM, Invitrogen Corporation, Carlsbad, CA, USA) supplemented with 10% fetal bovine serum (FBS) in 5% CO₂ at 37 °C. Cell viability was measured using a cell-counting kit-8 (AR1160, Boster Biological Engineering Co., Ltd., Wuhan, China).

2.3. Cell co-culture assay

To validate whether CRC cells induce endothelial apoptosis, a cell culture insert with 0.4 μm pore size was used as previously described [22]. The insert was inserted into a 6-well plate. A total of 20,000 HUVECs and 40,000 CT-26 tumor cells (suspended in DMEM with 3% FBS) were seeded onto the lower and upper chambers, respectively. After 24 h of culture, HUVECs were washed with phosphate-buffered saline (PBS) and fixed with 4% paraformaldehyde (dissolved in PBS). After washing again with PBS, HUVECs at the bottom of the plate were stained with hematoxylin and eosin. Images of endothelial morphology were recorded using a fluorescence microscope (Leica (Shanghai), Shanghai, China).

2.4. Western blotting assay

To detect the change in cleaved-polyadenosine diphosphate ribose polymerase (PARP), HUVECs were pretreated with vehicle, metformin (2 mM), 4-OOH-CTX (10 μM), or their combination for 4 h. Cells were digested with 0.25% trypsin and the protein concentration was determined using the bicinchoninic acid protein assay [23]. Protein of 80 μg was loaded into each lane and separated by 10% sodium dodecyl sulfate polyacrylamide gel electrophoresis. Subsequently, the protein was transferred onto a polyvinylidene fluoride membrane, which was then incubated with cleaved-PARP antibody (1:1500, #9548, Cell Signaling Technology, Boston, MA, USA) overnight. After incubation with an horseradish peroxidase (HRP)-conjugated secondary antibody (1:3000, #7076, Cell Signaling Technology), the final result was recorded using a chemiluminescence imaging system (Bio-Rad Laboratories, Hercules, CA, USA).

2.5. Xenograft experiments

All animal experiments were approved by the Animal Care and Use Committee of Xi'an Jiaotong University (No. 2016-249). BALB/c and nude mice were provided by the Animal Experiment Center of Xi'an Jiaotong University in a specific pathogen-free feeding environment. Briefly, 1 × 10⁶ murine 4T1 breast cancer cells were injected into the fat pad of female BALB/c mice, and CT-26, SW-480, and SW-620 CRC cells were injected into the rectal mucosa of BALB/c and nude mice. After seven days, mice were randomly divided into groups under different treatments (n = 8). To evaluate the chemosensitization effect, CT-26 tumor-bearing BALB/c mice were treated with metformin (225 mg/kg/day) for five days before intraperitoneal injection of CTX (20 mg/kg). The length and width of the transplanted tumors were measured every two to three days. Tumor volumes were calculated using the following formula: V = (length × width²)/2.

2.6. Lung metastasis model

To observe the non-angiogenic phenotype of CT-26 CRC cells growing in the lungs, 2 × 10⁵ CT-26 cancer cells suspended in 100 μL of PBS were intravenously injected into the tail vein of BALB/c mice. After 14 days, the mice were sacrificed by intraperitoneal administration of an overdose of pentobarbital, followed by an intracardiac injection of saline (50 mL). Finally, lung tissues were extracted and fixed for further experiments.

2.7. Immunohistochemistry (IHC) and hematoxylin and eosin (H&E) staining

For IHC and H&E staining, tissues were fixed with 10% formalin solution, embedded in paraffin, and cut into 4-μm sections as previously described [24]. To compare the morphology of the CD31⁺ vessels in various cancers, a standard immunohistochemical procedure was performed. The sections were dewaxed and hydrated, and the CD31 antigen was retrieved using Tris-ethylenediaminetetraacetic acid antigen retrieval solution (PH 9.0). After blocking with 5% bovine serum albumin (BSA) for 1 h at 37 °C, the sections were incubated with CD31 antibody (1:100, ab28364, Abcam (Shanghai), Shanghai, China), followed by HRP-conjugated secondary antibody (Boster Biological Engineering Co., Ltd.). Finally, the sections were washed with PBS, incubated with diaminobenzidine, and stained with hematoxylin. Images of whole sections were recorded using a slide scanner (Leica (Shanghai)).

2.8. Tumor hypoxia detection

Tumor hypoxia was detected as previously described [24]. Ninety minutes before the mice were sacrificed, 60 mg/kg pimonidazole (PIMO, Hypoxyprobe Inc., Burlington, Massachusetts, USA) was injected into the mice through the tail vein. After paraffin-embedding and section preparation, the tumor sections were immunostained with an anti-PIMO antibody to detect the percentage of PIMO⁺ hypoxic areas (percentage of total areas).

2.9. Immunofluorescence

For immunofluorescence imaging, tumor sections were prepared as previously described [24]. Briefly, tissues were fixed with 4% paraformaldehyde (12 h), sequentially hydrated, and embedded in optimal cutting temperature (OCT) compound. For 3D reconstruction of CD31⁺ vessels, tumor tissues were cut into 40 μm-thick sections, otherwise 6 μm-thick sections. Sections were permeabilized with 0.2% Triton-X 100 for 10 min and blocked with 5% BSA for 1 h at 37 °C. The following primary antibodies were used: CD31 (1:50, Ab28364; 1:40, ab7388; Abcam (Shanghai)), α-smooth muscle actin (α-SMA; 1:60, BM0002, Boster Biological Engineering, Co., Ltd.), NG-2 (1:80, R&D Systems, MAB6689, Minneapolis, MN, USA), vascular endothelial (VE)-cadherin (1:50, No. 138101, BioLegend (Beijing), Beijing, China), proliferating cell nuclear antigen (PCNA) (1:30, BM3888, Boster Biological Engineering, Co., Ltd.), cleaved (cl)-PARP (1:100, #9542, Cell Signaling Technology), cl-caspase (cl-Cas)-6 (1:100, #9761, Cell Signaling Technology), cl-Cas-7 (1:100, #8438, Cell Signaling Technology), cl-Cas-8 (1:100, #8592, Cell Signaling Technology), cl-Cas-9 (1:100, #9542, Cell Signaling Technology), and cisplatin-modified DNA antibody (GTX17412, GeneTex, Irvine, CA, USA). The secondary antibodies were 488-conjugated goat anti-rabbit antibody (1:150, A-11008, Invitrogen Corporation), 488-conjugated goat anti-rat antibody (1:150, A-11006, Invitrogen Corporation), 546-conjugated donkey anti-mouse antibody (1:200, A-10036, Invitrogen Corporation), and 647-conjugated donkey anti-rat antibody (1:150, A48272TR, Thermo Fisher Scientific Inc., Waltham, MA, USA). Fluorescent images were recorded using a confocal laser scanning microscope (Leica (Shanghai)). 3D-reconstruction of CD31⁺ vessels was performed using the LAS AF Lite software (Leica (Shanghai)). Tumor perfusion was analyzed as previously described [24].

2.10. Quantification of tissue hemoglobin

To quantify hemoglobin concentration, equal amounts of tumor tissue were suspended, sonicated with ultrasonic vibration equipment, mixed with Drabkin's reagent, and incubated in the dark at room temperature. The absorbance at 450 nm was recorded using a microplate reader (PerkinElmer, Waltham, MA, USA). Mouse hemoglobin was used to create a standard curve to determine the hemoglobin concentration in unknown samples.

2.11. Transcriptome sequencing assay

Transcriptome sequencing was performed as previously described [24]. HUVECs (1 × 10⁷) cultured with 2 mM metformin or vehicle were collected for quality control and sequencing (Biomarker Corporation, Beijing, China).

2.12. Statistical analysis

Quantitative data are presented as mean ± standard error of mean. Before analysis, the Kolmogorov-Smirnov normality test was performed to verify whether the data followed normal distribution. Statistical significance was calculated using a two-tailed Student's t-test or analysis of variance. Statistical significance was set at P < 0.05.

3. Results

3.1. CRCs were characterized by decreased MVD and reduced vascular CD31 expression

To validate whether the vasculature of CRCs is abnormal, the vascular morphologies of various cancers, organs, and normal tissues were compared. The vascular system in the human body consisted of a hierarchical network of normal tissues and organs (Fig. S1A). The vasculature of normal tissues consisted of arteries, veins, and capillaries with different morphologies. Normal tissues and organs with different functions and structures exhibited significant discrepancies in the MVD (Figs. S1A and B). Intriguingly, CT-26 CRC had a lower MVD than that of the rectal tissue (Figs. 1A, S1A and S1B). In addition, the vasculature of CT-26 cancer did not exhibit strong vascular sprouting or branching abilities (Fig. 1A). Unlike normal rectal tissue, vessels of CRCs were found to be poorly covered by vascular smooth muscle cells (VSMCs). These data suggest that the growth of CT-26 CRC was independent of angiogenesis, and that its vasculature was immature.

Fig. 1.

Vascular morphology, maturity, and leakage status of colorectal cancers. (A) Immunofluorescent double staining for CD31 (endothelia marker, green) and α-smooth muscle actin (α-SMA, vascular smooth muscle cell marker, red) of sections of murine colorectal CT-26 and the normal rectum. White and yellow triangles indicate vascular smooth muscle cells disassociated and associated with CD31⁺ vessels, respectively; white arrows indicate capillary-like vessels; M indicate mucosal layer of recta. Nuclei were counterstained with diamidinophenyl indole (DAPI). (B) Representative images of immunohistochemistry staining with antibodies against CD31 on paraffin embedded sections of murine normal rectum and CT-26, 4T1, and SW-480 cancers. Red and black arrows indicate CD31⁺ vessels with normal or decreased CD31 expression, indicating vessel regression occurred in colorectal cancers. (C) Double staining for CD31 and vascular endothelial (VE)-cadherin (marker for vascular basement membrane), revealing decreased the thinned vascular basement membrane in colorectal cancers. (D) Pimonidazole (PIMO; brown) staining and quantification of PIMO⁺ hypoxic area (red arrows) in CT-26 cancers (n = 8). (E) Representative images showing fluorescein isothiocyanate-conjugated dextran (70 kD) leakage from CD31-stained vessels in both peritumoral and intratumoral regions of CT-26 transplanted cancer. Quantitative data are indicated as mean ± standard error of the mean. PTR: peritumoral region; ITR: intratumoral region.

It has been previously reported that 4T1 metastatic breast cancer exhibits excessive angiogenesis [22,24,25]. The vasculature of both CT-26 and SW-480 CRCs was compared with that of normal rectum and 4T1 angiogenic cancer [24]. In addition, both cancers showed similar histological morphologies (Fig. S2A). IHC staining for CD31⁺ vessels showed that the vessels of 4T1 angiogenic cancer were angiogenic and dilated (Fig. 1B), which was consistent with previous results [26]. Strikingly, the vascular morphology of murine CT-26 was similar to that of human SW-480 CRCs and normal rectum. However, vessels of both CRCs showed decreased CD31 signal intensity (Fig. 1B), even in the absence of CD31⁺ EC of the vascular lumen. These data suggested that this vascular abnormality was widespread in CRCs.

3.2. CRCs were vascular immature and leaky, but not severely hypoxic

Previous studies have demonstrated that vascular function and maturation are closely associated with the cancer chemosensitivity [24]. This study focused on the vascular basement membrane, hypoxia, and leakage. Double staining for CD31 and VE-cadherin showed that vessels in the normal rectum adhered with a continuous and abundant basement membrane (Fig. 1C). Conversely, the vessels of both CT-26 and SW-480 cancers were surrounded by a thin but often discontinuous basement membrane. Further results of detection of tumor hypoxia showed that hypoxic areas of CT-26 CRCs were approximately 5.08% (percentage of total areas, Fig. 1D), which was significantly lower than that of 4T1 angiogenic cancer (percentage of hypoxic area: 28.75%) [24]. These results indicate that CRCs are not as hypoxic as angiogenic 4T1 breast cancers.

The immature vasculature is often accompanied by severe vascular leakage. Therefore, tumor-bearing mice were injected with 70 kD fluorescein isothiocyanate-conjugated dextran (FCD) before sacrifice, which could leak from immature vessels [24,27]. As expected, there was a lot of FCD leaking out of the vessels in the intra-tumoral region (ITR) of CT-26 cancer (Fig. 1E), while the vessels located in peritumoral region were not leaky.

3.3. ECs of CRCs were in a state of quiescence

Since vessels of CRC do not have an angiogenic phenotype, we next verified whether CRC is non-angiogenic. 3D reconstruction of the CD31⁺ vessels was performed to observe the continuity of the vascular structure. The vasculature of muscle was continuous and regular, whereas that of 4T1 cancer was excessively angiogenic and chaotic (Fig. 2A). In contrast, CT-26 and SW-480 cancers exhibited discontinuous and less angiogenic vasculature, with decreased MVD and CD31 expressions (Figs. 2B and C).

Fig. 2.

Colorectal cancer is non-angiogenic and has endothelial apoptosis. (A) 3D-reconstruction of CD31⁺ vascular morphology of normal muscle tissues, and 4T1, CT-26, and SW-480 cancers from untreated mice, showing that compared to breast cancer and normal tissue, the structure of vessel of colorectal cancer was discontinuous. Quantification of (B) microvessel density and (C) CD31 fluorescent intensity of vessels of muscle and cancers (n = 8). (D and E) Triple fluorescent staining for CD31, α-smooth muscle actin (α-SMA), and proliferating cell nuclear antigen (PCNA) of sections of normal rectum and 4T1 and CT-26 cancers, revealing that endothelial cells (ECs) of vessels of colorectal cancer were quiescent. White arrows indicate PCNA⁺ proliferating endothelial cells of vessels of 4T1 cancer. (F) Hematoxylin and eosin (H&E) staining for CT-26 tumor sections from untreated BALB/C mice. CT-26 tumor cells were invading the normal peritumoral tissues and pre-existing vessels are included into the tumor region. Black arrows indicate the peritumoral muscle tissues; yellow triangles indicate the co-opted vessels, which were wholly surrounded by CT-26 tumor cells; black triangles indicate the co-opted vessels at the invading edge. (G) Tail vein injection of CT-26-green fluorescent protein (GFP) colorectal cancer cells to establish a pulmonary metastasis model. Sections of murine lung were counterstained with diamidinophenyl indole (DAPI) and CD31 antibody. The pre-existed vessels (white arrows) in pulmonary interstitial tissue were being co-opted by the CT-26 tumor cells. (H and I) Quantification of percentage of polyadenosine diphosphate ribose polymerase (PARP)⁺ apoptotic ECs (white arrows) of CD31⁺ vessels, revealing that ECs of non-angiogenic CT-26 colorectal cancer were apoptotic (n = 8). Quantitative data are indicated as mean ± standard error of the mean. ^∗P < 0.05; ^∗∗P < 0.01; ^∗∗∗P < 0.001.

Further characterization of PCNA showed that there was a large number of PCNA⁺ ECs in 4T1 cancer, indicating that ECs of angiogenic cancer were proliferating (Fig. 2D, white arrows). Conversely, only a few ECs of CT-26 cancer were positively stained by the PCNA antibody (Fig. 2E), which was similar to that of the normal rectum. These data suggest that the growth of CRC was non-angiogenic and independent of angiogenesis.

3.4. CRCs co-opted pre-existing vessels and were with endothelial apoptosis

At the edge of CT-26 cancer, tumor cells showed a weak ability to invade the adjacent muscles, but penetrated the submucosal soft tissue (Fig. 2F). Critically, pre-existing vessels distributed in the submucosal layer were surrounded by tumor cells (yellow triangles) or included in the tumor region (black triangles). Importantly, in CT-26 lung metastases, the tumor vasculature closely resembled the vascular architecture of the normal lung (Fig. 2G), suggesting that CT-26 cancer co-opted pre-existing alveolar capillaries [10,12].

It has been reported that vessel regression was found in non-angiogenic tumors, which was independent of angiogenesis in ITR, from the middle stage but not the early stage [[28], [29], [30]]. Consistently, vessels in the core of CT-26 cancer were fragmented and expressed a high level of active PARP (Fig. 2H). In addition, 27.8% of CD31⁺ vessels were PARP⁺ in CT-26 cancer (Fig. 2I), whereas 4T1 cancer only had a low percentage (5.3%) of PARP⁺ vessels. These data may explain why the MVD of non-angiogenic CRC was lower than that of the original rectum.

3.5. CRCs were resistant to chemo-therapy via an in vivo mechanism

It has been reported that MVD and vascular maturity were closely associated with chemotherapy sensitivity [31]. Therefore, we investigated whether non-angiogenic CRC is resistant to chemotherapy, and selected SW-620 angiogenic cancer as a control. The viability of SW-480, SW-620, and CT-26 CRC cells was significantly inhibited by both 4-OOH-CTX and 5-FU in vitro (Figs. 3A–F). Consistently, when transplanted into mice, the growth of SW-620 and 4T1 angiogenic cancers was greatly suppressed by CTX administration (Figs. 3I and S2B). In contrast, non-angiogenic CRCs (SW-480 and CT-26) exhibited low sensitivity to CTX in vivo (Figs. 3G and H). The discrepancy between in vitro and in vivo chemosensitivity raised the question of whether the chemosensitivity of CRC is associated with an in vivo mechanism.

Fig. 3.

Non-angiogenic colorectal cancers were resistant to chemotherapeutic drugs in vivo. (A–F) In vitro proliferation curve of SW-480, CT-26, and SW-620 colorectal cancer cells treated with vehicle or 4-OOH-cyclophosphamide (CTX, active form of CTX) or 5-fluoro-pyrimidinedione (5-FU, repeated for 5 times) for 24 or 48 h. (G–I) Growth curve (cm³) of CT-26, SW-480, and SW-620 transplanted cancers from mice treated with vehicle or CTX at a dose 20 mg/kg (n = 8). Quantitative data are indicated as mean ± standard error of the mean. ns: no statistical significance. ^∗P < 0.05; ^∗∗P < 0.01; ^∗∗∗P < 0.001.

3.6. Metformin increased MVD and restored the consecutive vascular morphology in CT-26 non-angiogenic cancer

Since metformin has the potential to inhibit endothelial apoptosis [32]; therefore, we focused on the effect of metformin on the vasculature of non-angiogenic CRC. Metformin administration significantly increased the MVD of CT-26 cancer and restored vascular continuity, while having no effect on the sprouting and branching abilities of vessels (Figs. 4A and B). Since vessels derived from pro-angiogenic signaling-induced angiogenesis are characterized by apparent branching and sprouting, these data indicated that metformin-induced effects on MVD might not correlate with enhanced pro-angiogenic signaling. This notion was further validated by our finding that metformin did not alter the staining intensities of pro-angiogenic factors (Fig. S2C), including vascular endothelial growth factor (VEGF), platelet derived growth factor-B (PDGF-B), placental growth factor, and fibroblast growth factor-2 (FGF-2).

Fig. 4.

Effects of metformin on microvessel density and vascular morphology, maturity, and leakage of non-angiogenic cancer. (A) CD31 immunostaining in sections of CT-26 colorectal tumors from mice treated with vehicle or metformin (25 or 225 mg/kg/day). White arrows indicate capillary-like vessels; white triangles indicate vessels with low CD31 expression. (B) Quantification of vascular sprouts and branches (n = 8). (C) Representative images showing the diameter distribution of vessels in CT-26 tumors from vehicle- or metformin-treated mice (n = 8). Double immunostaining for (D) CD31/vascular endothelial (VE)-cadherin and (E) CD31/α-smooth muscle actin (α-SMA), showing enhanced vascular smooth muscle cell coverage and improved basement membrane in metformin-treated CT-26 tumors. White arrows indicate the VE-cadherin⁺ vascular basement membrane. (F) Quantification of the percentage of α-SMA⁺ vessels (indicated by CD31⁺ fluorescent signal, green) in CT-26 transplanted tumors treated with vehicle or metformin (n = 8). (G and H) Double immunostaining for CD31 (green) and NG-2 (red, pericyte marker), showing unaffected percentage of NG-2⁺ vessels and increased NG-2/CD31 ratio, NG-2⁺ pericyte areas, and percentage of pericyte association with vessels (PC association) in metformin-treated CT-26 tumors than control tumors (n = 8). (I and J) Representative images showing more lectin-perfused CD31⁺ vessels in metformin-treated than vehicle-treated CT-26 transplanted cancers (n = 8). Tetraethyl rhodamine isothiocyanate-conjugated lectin (red) was intravenously injected into tail veins 15 min before the sacrifice. Tumor sections were further immunostained with anti-CD31 antibody (green). White arrows indicate CD31⁺ vessels with blood perfusion. (K) Representative images showing fluorescein isothiocyanate-conjugated dextran (70 kD; white arrows) leakage from CD31⁺ vessels in vehicle- or metformin-treated CT-26 cancer. White triangles indicate dextran was leaking out of the vessel. Metformin administration reduced the vascular leakage in non-angiogenic CT-26 colorectal tumor. Quantitative data are indicated as mean ± standard error of the mean. ns: no statistical significance. ^∗P < 0.05; ^∗∗P < 0.01; ^∗∗∗P < 0.001. DAPI: diamidinophenyl indole.

Metformin also induced a shift of vessels towards a smaller size due to an increase in the number of capillary-like vessels (Fig. 4C, arrow). Importantly, metformin restored the non-continuous status of the vascular network of CT-26 cancer to a continuous vascular phenotype similar to that of skeletal muscle (Figs. S3A and B), which was accompanied by a reversal of CD31 expression in the vascular endothelium (Fig. 4A). Since endothelial apoptosis leads to reduced microvessels and diminished consecutive vasculature, the question arises as to whether or not metformin-induced increase in MVD and CD31 expression involves protection against apoptosis.

3.7. Metformin restored vascular consecutiveness, maturity, and functionality of non-angiogenic cancer

Induction of endothelial apoptosis leads to endothelial denudation (detachment from the smooth muscle layer) and vascular dysfunction [33,34]. This naturally raises the question of whether inhibition of endothelial apoptosis leads to restoration of vascular maturity and functionality in cancer. As shown in Fig. 4D, more abundant VE-cadherin deposition was found in the vessels of metformin-treated CT-26 non-angiogenic cancer. Strikingly, metformin significantly increased VSMCs and pericyte coverage (NG-2/CD31, Figs. 4E, G and H), but had no effect in elevating the percentages of VSMC⁺ and NG-2⁺ vessels (percentage of total vessels, Figs. 4F and H). Intravenous injection of lectin and FCD showed that metformin induced a significant increase in the density of lectin⁺ perfused vessels and led to a decrease in FCD leakage from the vessels (Figs. 4I and K). Overall, these data indicated that metformin restored vascular consecutiveness, maturity, and functionality in non-angiogenic CRC.

3.8. Metformin-induced chemosensitization depended upon a vascular mechanism

As improved vascular maturation and function allows more chemo-drugs to be delivered into cancer [8,27], we then investigated whether metformin pretreatment sensitizes cancer cells to CTX (20 mg/kg/day). Compared with CTX treatment alone, metformin pretreatment significantly enhanced the anticancer effect of CTX and prolonged the survival of tumor-bearing mice (Figs. 5A and B). This effect was accompanied by a significant increase in tumor necrosis and hemorrhage (Figs. 5C and D). Further characterization of the pattern of necrosis and hemorrhage showed that a fraction of microvessels was occluded by thrombus in untreated CT-26 cancer, which was indicated by a mass of red blood cells gathering in the intravascular lumen in a funicular shape (Fig. 5C, yellow triangles). The results of IHC staining showed that more cisplatin-DNA adduct-positive cells were found in metformin-pretreated CT-26 cancer (Fig. S3C). These data suggest that metformin pretreatment had the potential to enhance the toxicity of CRC to chemotherapies.

Fig. 5.

Metformin induced chemosensitization in non-angiogenic colorectal cancer via a vascular mechanism. (A) Growth curve and (B) survival curve of CT-26 colorectal cancers from mice treated with vehicle, metformin (225 mg/kg), low dose cyclophosphamide (CTX, 20 mg/kg), or the combined treatment (n = 8). Mice were pretreated with metformin for 5 days before CTX was given. (C) Hematoxylin and eosin staining of sections of CT-26 cancers, and (D) quantification of necrotic and hemorrhagic areas revealed enhanced cancer responses to CTX by metformin pretreatment (n = 8). Black arrows indicate tumor necrosis; yellow triangles indicate microvessel occlusion adjacent to the necrotic region. Magnification: 200×. (E) Representative images showing iron containing hemoglobin distributed in CT-26 cancers. There appears to be more hemoglobin in deep regions of CT-26 tumors from mice with the combined metformin (pretreatment for 5 days) and CTX treatment. Black arrows indicate the hemoglobin that gathered in the marginal region; blue arrows indicate the hemoglobin that existed in the deep region. (F) Quantification of hemoglobin (Hb) content in CT-26 tumors of different groups (n = 8–18). (G) Micrographs of fluorescein isothiocyanate-conjugated dextran (FCD)-perfused and CD31-immunostained vessels in control- and metformin-treated CT-26 cancers, revealing enhanced vascular leakage by metformin pretreatment. Tumor-bearing mice were orally pretreated by metformin with allowance for restoration of vascular function. White arrows indicate FCD leaking outside the vessel lumen. (H) Double staining for CD31/cleaved-caspase-9 and (I) quantification of cleaved-caspase-9⁺ CD31⁺ cancer cells and vessels in CT-26 cancer (n = 8). Tumor-bearing mice were treated with metformin (pretreatment for 5 days), CTX, or combined treatment. White arrows indicate caspase-9⁺ endothelial cell (EC); yellow triangles indicate caspase-9⁺ cancer cells. (J) Immunoblotting for both inactive and cleaved-polyadenosine diphosphate ribose polymerase (PARP) in ECs treated with vehicle, 100 μM metformin, 10 μM 4-OOH-CTX, or the combined treatment for 24 h, and (K) quantification of the ratio of cleaved-PARP/PARP (n = 6). (L and M) Quantification of propidium iodide (PI)⁺ necrotic CT-26 colorectal cancer cells treated with 4-OOH-CTX (activated CTX), metformin, or the combined treatment (n = 5). Metformin had no direct effect on the sensitivity of CT-26 cells to 4-OOH-CTX in vitro. Quantitative data are indicated as mean ± standard error of the mean. ns: no statistical significance. ^∗P < 0.05; ^∗∗P < 0.01; ^∗∗∗P < 0.001.

Importantly, metformin pretreatment significantly enhanced CTX-induced promotion of vessel occlusion (Figs. 5C and D), thus increasing tumor necrosis. Direct observation and hemoglobin detection using Drabkin's reagent showed that the hemoglobin content of reticulocytes was significantly increased by CTX in CT-26 cancer when mice were pretreated with metformin (Figs. 5E and F). Interestingly, the combined therapies enhanced vascular leakage (Fig. 5G). This pattern was very similar to vascular disrupting agents-induced tumor necrosis [35], which is secondary to endothelial apoptosis-associated vascular dysfunction. As endothelial apoptosis leads to occlusion of microvessels, metformin-induced prevention of vessel occlusion (Fig. 5C) might result from the inhibition of endothelial apoptosis.

Furthermore, more cl-Cas-9⁺ apoptotic cells and vessels were found in CT-26 cancers from mice treated with the combination therapies (Figs. 5H and I). Interestingly, apoptotic cells were mainly concentrated in the area near blood vessels (Fig. 5H). However, metformin pretreatment significantly increased the number of cl-Cas-9⁺ vessels in CT-26 cancer, while having no effect on 4T1 angiogenic cancer [24]. Since CTX cannot be directly activated in vitro, we used 4-OOH-CTX, the activated isoform of CTX, for further observation of the in vitro effects of CTX on apoptosis of cancer and endothelial cells. Critically, metformin did not affect CTX-induced cellular apoptosis and propidium iodide (PI)⁺ necrosis in vitro (Figs. 5J and M). Overall, metformin-induced chemosensitization might result from enhanced drug delivery via a vascular mechanism rather than a direct enhancement of cancer cell sensitivity.

3.9. Metformin did not promote endothelial proliferation

As metformin increased MVD in CRC, we next explored whether the effect is dependent on promoting endothelial proliferation. Sections of both CT-26 and SW-480 CRCs were stained with CD31 and PCNA antibodies (Figs. 6A and C). As shown, the percentage of vessels with PCNA⁺ EC in CT-26 and SW-480 cancers was 6.66% and 8.69% (Figs. 6A–D), respectively, which were significantly lower than that of 4T1 angiogenic cancer (Fig. 2E). This phenomenon was further verified by the fact that few vessels had apparent vascular sprouting. Consistently, the percentage of vessels with PCNA⁺ ECs in both cancers was not affected by metformin administration (Figs. 6B and D). Further results of RNA sequencing showed that the expression of genes associated with cell cycle, apoptosis, proliferation, and angiogenesis of HUVECs was not apparently affected by metformin in vitro (Figs. 6E–G). These data suggested that the metformin-induced elevation of MVD was not dependent on the promotion of endothelial proliferation.

Fig. 6.

Effect of metformin on endothelial proliferation of colorectal cancers. Mice bearing SW-480 or CT-26 colorectal cancer cells were administered with vehicle or metformin. Double immunofluorescent staining for CD31 (endothelial marker, red) and proliferating cell nuclear antigen (PCNA, green), revealing unaffected proliferation state of endothelial cells of (A) CT-26 and (C) SW-480 colorectal cancers. White arrows indicate PCNA⁺ proliferating endothelial cell. Quantification of vessels with proliferating endothelial cell of (B) CT-26 and (D) SW-480 colorectal cancers, which was identified by CD31⁺ cells with PCNA positivity in the nucleus (n = 8). Heatmap analysis of mRNA levels of (E) cell cycle, proliferation, (F) apoptosis, and (G) angiogenesis-associated genes of human umbilical vascular endothelial cells (HUVECs) treated with vehicle or 2 mM metformin (three independent samples). All those genes were not significantly affected by metformin. ns: no statistical significance between groups; Rep: report; CCND: cyclin D; CCNE: cyclin E; CDK: cyclin dependent kinase; Bad: BCL2 associated agonist of cell death; BID: BH3 interacting domain death agonist; Cas: caspase; ENG: endoglin; FGF-2: fibroblast growth factor 2; HGF: hepatocyte growth factor; HIF: hypoxia inducible factor; PDGF-A: platelet derived growth factor A; PDGF-B: platelet derived growth factor B; PIGF: placental growth factor; VEGF-A: vascular endothelial growth factor A; VEGF-C: vascular endothelial growth factor C; MMP: matrix metallopeptidase; IGFBP: insulin like growth factor binding protein.

3.10. Caspase-mediated endothelial apoptosis was inhibited by metformin

As increased MVD was not mediated by the promotion of endothelial proliferation, we explored whether endothelial apoptosis is involved in this mechanism. The results of fluorescent staining showed that the CD31⁺ vessels of CT-26 cancer were not positive for cl-Cas-6 or cl-Cas-7 (Figs. 7A and B). However, the vessels were positively stained for cl-Cas-9 and cleaved-PARP (Figs. 7C and D), which was accompanied by decreased CD31 expression. Compared with vehicle-treated cancer, metformin-treated CT-26 cancer had a significantly lower percentage of caspase-9⁺ and PARP⁺ vessels (Figs. 7E and F). Since vascular apoptosis was accompanied by a decrease in endothelial CD31 expression (Fig. 7G), the metformin-induced increase in endothelial CD31 expression suggested amelioration of endothelial apoptosis (Fig. 7H).

Fig. 7.

Metformin inhibited endothelial apoptosis of non-angiogenic cancer. CT-26 colorectal cancer cells were injected into the rectum mucosa of BALB/C mice to establish an orthotopic transplantation tumor model. Double staining for CD31 and (A) cleaved-caspase (cl-Cas)-6 or (B) cl-Cas-7 of CT-26 tumor, indicating that endothelial apoptosis in non-angiogenic CT-26 cancer is independent of Cas-6 and Cas-7 signaling. CD31⁺ endothelial cells are not positive for cl-Cas-6⁺ and cl-Cas-7⁺. Double staining for CD31 and (C) cl-Cas-9 or (D) active polyadenosine diphosphate ribose polymerase (PARP) of CT-26 tumors from vehicle- or metformin-treated mice. White triangles indicate vessels with non-apoptotic endothelial cells; yellow triangles indicate apoptotic endothelial cells with decreased CD31 expression. Quantification of (E) cl-Cas-9⁺ tumor cells per field and (F) percentage of vessels with cleaved-caspalse-9⁺ or active PARP⁺ endothelial cells (percentage of CD31⁺ vessels) in CT-26 tumors (n = 8). (G) Spearman analysis of linear correlation between cl-Cas-9 intensity and CD31 intensity of vessels in CT-26 tumors (n = 8). CD31 intensity of vessels is negatively correlated with cl-Cas-9 intensity. (H) Quantification of mean fluorescent intensity of CD31 in vessels of vehicle- or metformin-treated CT-26 tumors (n = 8). (I) Triple fluorescent staining for CD31 (purple), α-SMA (yellow), and cl-Cas-3 (blue) of sections of human colorectal cancer and peri-tumoral normal rectal tissue. Compared to normal recta, vessels of colorectal cancer had endothelial apoptosis (white triangles). Quantitative data are indicated as mean ± standard error of the mean. ns: no statistical significance. ^∗P < 0.05; ^∗∗P < 0.01; DAPI: diamidinophenyl indole.

We next investigated whether this phenomenon also occurs in human CRCs. Sections of human CRCs and peritumoral tissues were stained for CD31/α-SMA/cl-Cas-3, which is a critical executioner of apoptosis. In peritumoral tissues, there were many vessels with a well-structured lumen, adequate pericyte coverage, and critical non-endothelial apoptosis (Fig. 7I). However, a large number of vessels undergoing endothelial apoptosis and subsequent structural defects were found in CRC. Consistently, those cl-Cas-3⁺ endothelial cells also showed decreased CD31 expression. Therefore, these results suggested that endothelial apoptosis-associated vascular abnormalities might be a widespread phenomenon.

3.11. Metformin suppressed CRC-induced endothelial apoptosis

Because tumor cells can affect the status of ECs in microenvironment [27], we explored whether CRCs directly mediate endothelial apoptosis, and whether this mechanism is prevented by metformin. Consistent with our previous results [22], 4T1 cancer cells, which had an angiogenic phenotype in the transplanted model, promoted endothelial sprouting and activation (Fig. 8A). In contrast, when co-cultured with CT-26 CRC cells in a co-culture system [22], HUVECs underwent apoptosis and cell death. Strikingly, the CRC cell-mediated effect was suppressed by metformin (Fig. 8A), which was consistent with its in vivo effect. Overall, these data suggested that CRC-induced endothelial apoptosis could be abrogated by metformin treatment.

Fig. 8.

Effect of metformin on colorectal cancer (CRC) cell-induced endothelial apoptosis. (A) Representative images for hematoxylin and eosin staining of human umbilical vascular endothelial cells (HUVECs) co-cultured with 4T1 and CT-26 cancer cells treated with vehicle or metformin (2 mM). Metformin abrogated CT-26 cancer cells co-culture-induced cell death of HUVECs. Red triangles indicate endothelial vascular sprouts; red arrows indicate cell fragments. Magnification: 200×. (B) Schematic diagram of chemosensitization-mediated by metformin in non-angiogenic CRC. Compared to angiogenic metastatic breast cancers, CRCs are independent of angiogenesis and characterized by endothelial apoptosis and pericyte coverage loss, which lead to vascular immaturity, leakage, and chemoresistance. By suppressing endothelial apoptosis, metformin increased the microvessel density and improved the vacular maturity, therefore enhancing drug delivery and inducing chemosensitization in non-angiogenic CRCs.

4. Discussion

To date, it remains unclear whether it is worth avoiding vessel regression or EC apoptosis in the middle-late stage of non-angiogenic cancer. In the current study, endothelial apoptosis induced an immature and dysfunctional vascular network, thus limiting chemo-drug access to the tumor core. This speculation is supported by the fact that non-angiogenic CRC showed a worse response to CTX treatment in vivo than in vitro. By inhibiting endothelial apoptosis, metformin increased MVD and restored the vascular structure, function, and maturity of non-angiogenic CRCs, thus inducing chemosensitization (Fig. 8B). Seemingly, increased MVD might accelerate tumor growth, but it might allow more metformin and immune cells to enter the tumor.

It has been widely accepted that tumor growth strictly depends on angiogenesis, formation of new blood vessels from the pre-existing ones [36,37]. However, there is a growing body of evidence to support the fact that tumors can also grow without the induction of angiogenesis by co-opting pre-existing vessels [9,10]. Based on this evidence, a new aspect of interaction between tumor cells and ECs should be highlighted. Further efforts are needed to understand the biology of this previously unrecognized group of tumors. The status of ECs can be affected by tumor cells or immune cells in the microenvironment [38,39]. This naturally raises a question of whether or how the status and function of vessels are affected by non-angiogenic cancer after co-option. Herein, our results showed that endothelial apoptosis-mediated by caspases signaling occurred in the vessels of non-angiogenic CRCs, accompanied by reduced MVD and vascular immaturity and dysfunction.

The main reason for the poor prognosis of malignancies is that molecular mechanism underlying chemoresistance remains poorly understood. Over the past decade, metformin has been shown to sensitize in various cancers [[40], [41], [42]]. These studies have mainly focused on the direct effect of metformin at concentrations higher than the blood concentrations in mice (50–200 μM). Therefore, it is important to elucidate the underlying mechanisms, especially the in vivo mechanism. We previously demonstrated an in vivo mechanism of metformin, which involves targeting the tumor vasculature. Metformin pretreatment sensitizes transplanted breast cancers of angiogenic phenotype to chemotherapy [24]. In the current study, we reported a similar sensitization effect in CRC models, but with a different vascular phenotype. These data demonstrated that metformin had the potential to target different vascular abnormalities. Critically, increased MVD allows more chemo-drugs to be delivered into the tumor, thus ameliorating or abrogating chemoresistance [36].

Non-angiogenic cancers are resistant to anti-angiogenic therapies [11,43,44]. By co-opting pre-existing mature vessels, non-angiogenic tumors can survive and facilitate their growth independently of angiogenesis [45,46]. Recently, clinical evidence has demonstrated that non-angiogenic tumors respond poorly to anti-angiogenic therapies [11]. In preclinical studies [43,44], it was reported that the resistance of non-angiogenic tumors to anti-angiogenic therapies was mediated by vessel co-option. Biomarker or phenomenon that has a predictive value for response to anti-angiogenic therapies remains elusive [11]. Whether to treat non-angiogenic cancer with anti-angiogenesis therapies remains questionable. Thus, there is a pressing need to develop novel treatment strategies that can benefit patients with non-angiogenic cancers by affecting tumor vasculature. Our data suggested that protection against endothelial apoptosis may benefit patients with non-angiogenic CRCs, which needs to be further verified.

Although metformin significantly increased the MVD of non-angiogenic CRC, the tumor growth was not greatly promoted. The mechanism of this effect differs from that underlying metformin's anti-angiogenic effect in angiogenic cancer, which depends on suppression of pro-angiogenic factor expression [47,48]. Metformin administration did not increase the expression of VEGF, FGF-2, or PDGF-B, indicating that this effect was independent of angiogenesis induction. In contrast, the metformin-induced increase of MVD was mediated by targeting endothelial apoptosis, which reduced the number of vessels co-opted by tumor cells. Increased blood flow allows more cytotoxic T cells to enter the tumor region [49], thus partially or completely abrogating MVD elevation-induced tumor growth promotion. Therefore, these results suggest that targeting endothelial apoptosis-induced MVD elevation may have different implications from those mediated by pro-angiogenic factors, which warrant further attention.

The non-angiogenic phenotype of CRC has both similarities and differences with the results reported recently in models of liver or lung metastasis [10,15,29]. After transplantation into the rectal mucosa, vessels in CT-26 CRC had a low rate of PCNA⁺ ECs and no apparent vascular sprout. This phenomenon indicates that ECs are not activated to induce angiogenesis in an in situ model of CRC. In the edge region, tumor cells are incorporated pre-existing vessels and soft tissue, but not muscle. This vessel co-option phenotype is observed in an orthotopic tumor. A similar vascular phenotype was observed in SW-480 xenografts, established in nude mice. This evidence suggests that this vascular phenotype may not be affected by T cell immunity [43].

The anti-apoptotic effects of metformin have also been demonstrated in other pathological models. Several studies have reported that hyperglycemia induces endothelial apoptosis and dysfunction through its major contributor, oxidative stress. Tumor cells induce necroptosis in ECs, which promotes tumor cell metastasis [16], suggesting that both substances and cells can cause EC death. Eriksson and Nyström [18] found that metformin protects human coronary artery endothelial cells against diabetic lipoapoptosis by activating AMPK. Furthermore, metformin prevented EC death by suppressing the mitochondrial permeability transition pore, which is a mitochondrial channel. Therefore, our results are consistent with the evidence that metformin has the potential to protect EC against cell death and apoptosis under different pathological conditions.

5. Conclusions

Our current work provides evidence for the vascular mechanism underlying metformin-induced chemosensitization of CRCs with a non-angiogenic phenotype. Endothelial apoptosis induced by CRCs might also have effects on vascular maturity, function, and perfusion, which is critical for chemo-drug delivery. Thus, targeting the apoptosis of ECs of CRCs may provide a strategy for overcoming chemoresistance.

CRediT author statement

Guang-Yue Li and Shu-Jing Zhang: Writing - Original draft preparation, Reviewing and Editing, Conceptualization, Methodology; Dong Xue and Yue-Qi Feng: Data curation, Writing - Reviewing and Editing; Yan Li and Xun Huang: Funding acquisition, Resources; Qiang Cui and Bo Wang: Formal analysis; Jun Feng: Funding acquisition, Validation; Tao Bao and Pei-Jun Liu: Validation, Conceptualization; Shao-Ying Lu and Ji-Chang Wang: Writing - Original draft preparation, Conceptualization, Methodology, Funding acquisition.

Declaration of competing interest

The authors declare that there are no conflicts of interest.

Appendix A. Supplementary data

The following is the Supplementary data to this article.