Deciphering Colorectal Liver Metastasis Perfusion Heterogeneity: Identification of Distinct Vascular Phenotypes through High-Lateral-Resolution Photoacoustic Microscopy

Abstract

Purpose

To develop a photoacoustic microscopy (PAM) method for classifying colorectal liver metastasis (CRLM) microvascular networks as angiogenesis-type or vessel co-option–type at different metastatic stages and to evaluate their responses to bevacizumab.

Materials and Methods

Seventy-two BALB/c nude mice received intrasplenic injections of HCT116-luc or HT29-luc cells to establish CRLM models. PAM was used to visualize CRLM vascular architecture, and its diagnostic accuracy was validated through histopathologic analysis. Intergroup differences in PAM parameters were assessed with one-way analysis of variance or independent-samples t tests.

Results

PAM differentiated angiogenesis-type from vessel-co-option–type CRLMs according to distinct vascular architectures: Angiogenesis-type CRLMs exhibited lower vessel density (15.66% ± 4.38 vs 31.44% ± 6.95; P < .001) and vessel junction density (1.19 × 10^-3 n/px² ± 0.52 vs 3.12 × 10^-3 n/px² ± 1.11; P = .001) and higher vessel diameter (31.45 µm ± 4.59 vs 13.68 µm ± 1.61; P < .001) and lacunarity (0.294 arbitrary units [au] ± 0.095 vs 0.148 au ± 0.059; P = .003) than vessel co-option–type CRLMs. These differences persisted throughout the 8-week progression period. A 33.3% reduction in vessel density at 3 weeks following bevacizumab treatment was observed in angiogenesis-type CRLMs (untreated, 15.66% ± 4.38 vs treated at 3 weeks, 10.44% ± 4.91; P = .047), with no evidence of a difference in vessel co-option–type CRLMs (untreated, 31.44% ± 6.95 vs treated at 3 weeks, 28.82% ± 3.56; P = .56).

Conclusion

PAM quantified microvascular signatures differentiating CRLM blood supply patterns and identified vessel co-option–type metastases as resistant to bevacizumab.

Keywords: Angiogenesis, Bevacizumab, Colorectal Cancer, Histopathologic Growth Pattern, Liver Metastases, Photoacoustic Imaging, Photoacoustic Microscopy, Vessel Co-Option

Supplemental material is available for this article.

© RSNA, 2026

See also editorial by Wang and Chen in this issue.

See also commentary by Haddad and Bhutiani in this issue.

Summary

A photoacoustic microscopy method combining quantitative parameters and vascular morphologic analysis from in vivo scanning was developed; it effectively distinguished colorectal liver metastasis vascular subtypes and identified subtypes that were resistant to bevacizumab treatment.

Key Points

• ■ A photoacoustic microscopy (PAM) method was developed to identify distinct blood supply patterns in colorectal liver metastases (CRLMs) in model mice.

• ■ Angiogenesis-type CRLMs showed lower vessel density (15.66% ± 4.38 vs 31.44% ± 6.95; P < .001) and vessel junction density (1.19 × 10^-3 n/px² ± 0.52 vs 3.12 × 10^-3 n/px² ± 1.11; P = .001) and higher vessel diameter (31.45 µm ± 4.59 vs 13.68 µm ± 1.61; P < .001) and lacunarity (0.294 arbitrary units [au] ± 0.095 vs 0.148 au ± 0.059; P = .003) than vessel co-option–type CRLMs.

• ■ A 33.3% reduction in vessel density at 3 weeks following bevacizumab treatment was observed in angiogenesis-type CRLMs (P = .047), with no evidence of a difference in vessel co-option–type CRLMs (P = .56).

Introduction

Colorectal cancer (CRC) is the third most prevalent malignancy worldwide and the second leading cause of cancer-related mortality (1). Over half of patients diagnosed with CRC develop liver metastases, which are responsible for more than 70% of CRC-associated deaths (1,2). Currently, systemic treatment of unresectable colorectal liver metastases (hereafter, CRLMs) involves oxaliplatin- and irinotecan-based chemotherapy regimens combined with targeted therapies such as bevacizumab (3,4). However, findings from comparative studies suggest that a subset of patients receiving bevacizumab plus chemotherapy exhibit lower survival rates than those receiving chemotherapy alone, with only a fraction deriving substantial benefit from the combination therapy (5,6). Frentzas et al (5) confirmed that these patients did not respond well to bevacizumab treatment because their tumors use vessel co-option rather than angiogenesis. Identification of these distinct blood supply patterns is essential for treatment and impacts clinical decision-making.

Until recently, the mechanism of CRLM blood supply could be deduced only by analyzing their histopathologic growth patterns (HGPs) (5,7,8) (Fig 1A). CRLMs with pushing and desmoplastic HGPs, reflecting angiogenesis, are sensitive to bevacizumab treatment, whereas those with replacement HGPs, characterized by vessel co-option, exhibit bevacizumab resistance (5,8,9). Studies have confirmed that there is a stable association between specific cell lines and HGP phenotypes. For example, Fleten et al (10) demonstrated in 2017 that HCT116 cells primarily produce liver metastases with a pushing HGP, whereas HT29 cells lead to lesions presenting a replacement HGP. This finding has also been supported by recent research. For example, Rada et al (11) similarly observed in 2021 that the HT29 cells almost exclusively produced metastases exhibiting a replacement HGP, indicating the blood supply pattern is mediated by vessel co-option. Clinically, imaging techniques such as CT and MRI are primarily used to assess tumor boundary and volume (12), with HGP determination requiring invasive biopsy for pathology analysis. Indeed, earlier studies attempted to predict the dominant HGP in liver metastases via radiomics, but to our knowledge, no minimally invasive or noninvasive techniques have been reported to directly visualize the intratumoral microvascular architecture (13,14).

Figure 1:

The association between blood supply pattern and response to bevacizumab (Bev) therapy in patients with colorectal liver metastases (CRLMs). (A) Blood supply patterns in CRLMs: (i) normal liver tissue, (ii) tumor–liver interface demonstrating vessel co-option, and (iii, iv) tumor–liver interface showing angiogenesis-dependent perfusion. (B) Configuration of the photoacoustic microscopy system used in this work. DAS = data acquisition system, FPGA = field-programmable gate array.

Photoacoustic imaging (PAI) is a novel imaging modality based on the photoacoustic effect: When biologic tissues are irradiated by short-pulsed laser beams, the target region absorbing optical energy undergoes transient thermoelastic expansion, generating detectable ultrasonic waves (15,16). PAI holds great promise in CRLM assessment because it uses nonionizing radiation to image tissue with a high spatial and contrast resolution; in addition, PAI provides real-time three-dimensional imaging capability and is associated with low costs and little patient discomfort (17). These advantages enable more precise characterization of the distribution patterns of intratumoral microvascular networks.

By using photoacoustic microscopy (PAM) with micrometer-scale resolution, we previously achieved synchronous three-dimensional quantification of intricate intratumoral microvascular networks (18) and obtained microvascular metrics not attainable through conventional methods. Based on this previous research, the aims of this study were to develop a PAM imaging method for the precise classification of CRLM intratumoral microvascular networks as angiogenesis-type or vessel co-option–type at the early and advanced metastatic stages and to evaluate the responses of these CRLM subtypes to antiangiogenic therapy with bevacizumab.

Materials and Methods

Cell Lines

The human CRC cell lines HCT116-luc and HT29-luc used in this study were obtained from Immocell Biotechnology. These cell lines were cultured in their respective dedicated media (McCoy 5a supplemented with 10% fetal bovine serum, 1% penicillin-streptomycin, and 1.0 μg/mL puromycin, purchased from Immocell) and maintained in an incubator at 37°C with 5% CO₂. The medium was replaced every 48 hours. The cells were maintained as adherent monolayers. When cell confluence reached approximately 80%–90%, the cells were detached using 0.25% trypsin solution (Pricella Biotechnology) and passaged (subcultured) at a ratio of 1:3. All cell lines were authenticated by short tandem repeat profiling to confirm the absence of cross-contamination with other human cell lines. Additionally, mycoplasma contamination was assessed using a mycoplasma detection kit (Immocell), and all tested cell lines were confirmed to be negative for mycoplasma.

CRLM Models

A total of 72 6-week-old male BALB/c nude mice (body weight, 20 g ± 2.5) were randomly divided into nine groups (n = 8 per group) and housed under specific pathogen-free conditions at 26°C. All animals were purchased from Bestest Biotechnology (RRID:IMSR_RJ:BALB-C-NUDE). The experimental protocol was approved by the ethics committee of Optoelectronic Science and Engineering, South China Normal University (approval no. SCNU-BIP-2025-056) and complied with all relevant ethics standards, including the guidelines of the institutional animal care and use committee.

Before the procedure, the surgical site and all surgical instruments were sterilized. Mice were anesthetized with isoflurane (Yuyan Instruments) and restrained in a lateral position. A 5-mm incision was made in the left abdominal flank to expose the spleen. The control group received 50 μL of physiologic saline. Using the HT29-luc cell line to establish the vessel co-option model (10,11), cells were resuspended in Matrigel (Beyotime Biotechnology ) (4 × 10⁶ cells/mL). Thirty-two mice were subjected to intrasplenic injection of tumor cells (50 μL containing 2 × 10⁵ cells). After tumor confirmation by bioluminescence imaging, mice were stratified into four subgroups (n = 8 per group) according to treatment status and PAM scanning time points: (a) baseline observation group, (b) tumor progression group, (c) 1-week bevacizumab treatment group, and (d) 3-week bevacizumab treatment group. Another 32 mice were inoculated with the HCT116-luc cell line to establish the angiogenesis model, following the same protocol (10). The complete experimental design and grouping strategy are summarized in the Table.

Study Design: Experimental Groups and Treatment Regimens

Group	No. of Mice	Injection Material	Subgroup Purpose	Treatment
1	8	HCT116-luc	Baseline observation	None
2	8	HCT116-luc	Tumor progression control	None
3	8	HCT116-luc	1-week therapy response	Bevacizumab (1 week)
4	8	HCT116-luc	3-week therapy response	Bevacizumab (3 weeks)
5	8	HT29-luc	Baseline observation	None
6	8	HT29-luc	Tumor progression control	None
7	8	HT29-luc	1-week therapy response	Bevacizumab (1 week)
8	8	HT29-luc	3-week therapy response	Bevacizumab (3 weeks)
9	8	Saline	Normal liver control	None

Note.—Bevacizumab was administered at a dose of 10 mg/kg via intraperitoneal injection twice weekly for the indicated duration.

PAI System

A schematic illustration of the PAM system used in this study is shown in Figure 1B. A pulsed Nd:YAG laser (DTL-319QT; Laser-export) served as the photoacoustic excitation source. This source emitted at a wavelength of 532 nm, which lies within the high optical absorption band of hemoglobin (for both oxygenated and deoxygenated forms), thereby enabling label-free contrast for microvasculature imaging (19,20). The laser pulse width was 10 nsec, with a pulse energy of approximately 200 nJ at the sample surface. The laser beam was expanded and filtered before being coupled into a single-mode fiber via a fiber coupler (PAF-X-15-PC-A; Thorlabs). System synchronization was controlled by a field-programmable gate array, which generated trigger signals to coordinate the laser and data acquisition system at a maximum repetition rate of 10 kHz.

The generated photoacoustic signals were first amplified by a broadband low-noise amplifier (PE15A63014; Pasternack) with a 60-dB gain, followed by digitization and storage with a single-channel high-speed data acquisition card (M4i.4420; Spectrum) at a sampling rate of 200 million samples per second. Imaging was performed with an ultrasonic transducer with a center frequency of 50 MHz, achieving an optical resolution of approximately 5 µm and a maximum detection depth of 1.0 mm.

Experimental Procedure

Longitudinal tumor monitoring via bioluminescence imaging

The tumor cells used in this study stably expressed a luciferase reporter gene. We monitored tumor growth weekly with a Clinx IVScope 8800 Small Animal In Vivo Imaging System (Clinx Science Instruments) starting from day 7 after injection. The purpose of monitoring was to confirm tumor growth. Once a positive bioluminescent signal was detected in a mouse, subsequent bioluminescence imaging for that animal was discontinued. Mice were anesthetized with isoflurane and injected intraperitoneally with the luciferase substrate d-luciferin (150 mg/kg; Aladdin) 10 minutes before imaging. Ten minutes after injection, mice were placed in the imaging chamber, and signals were acquired in complete darkness in bioluminescence mode. The system was configured with no excitation light and with an open filter to maximize signal capture. A fixed exposure time, which was initially optimized on the basis of signal intensity, was used consistently for all imaging sessions. We quantified the tumor signal by analyzing images with Clinx IVScopeEQ Capture software (version 4.10.5). A region of interest was drawn around the tumor, and its total radiant flux was measured.

PAM imaging of liver tissue and tumors

To acquire stable, high-lateral-resolution microvascular images, it was necessary to overcome two primary technical obstacles: signal attenuation due to the abdominal wall exceeding the system’s imaging depth and respiratory motion artifacts. To this end, before imaging, experimental mice were anesthetized with isoflurane, and a small abdominal incision was made to expose the liver. A thin plastic film and physiologic saline (0.9% NaCl) were used as coupling media for laser transmission and acoustic wave propagation. The imaging system provided a lateral resolution of 5 μm and a B-scan frame rate of 4 Hz. The spatial sampling step (ie, pixel size) of the system was set as follows: 1.0 μm along the fast axis and 0.5 μm along the slow axis. The maximum single-scan field of view was 4 mm × 4 mm (adjustable). Under this configuration, the effective imaging depth was approximately 0.6 mm.

The original acquired data were subsequently input into LabVIEW (National Instruments; RRID:SCR_014325) and MATLAB version R2023b (MathWorks; RRID:SCR_001622) to reconstruct the liver sinusoidal microvasculature structure and obtain multiple quantitative parameters. ImageJ (National Institutes of Health; RRID:SCR_003070) and AngioTool (21,22) software were used to visualize vascular segmentation and quantitatively analyze vessel characteristics (Fig 2B–2C; Appendix S1).

Figure 2:

Photoacoustic microscopy (PAM) imaging and quantitative analyses of microvascular features in normal liver tissue and two tumor subtypes. (A) Raw PAM scans of normal liver tissue and of tumors exhibiting distinct vascular phenotypes; images highlight tumor microvasculature. (B) Binarized versions of the PAM scans shown in A. (C) Vascular networks processed by AngioTool software (21,22) using images from B. (D–G) Quantitative comparisons include (D) vessel density, (E) vessel diameter, (F) lacunarity, and (G) vessel junction density across normal liver, angiogenesis-type tumors, and vessel co-option–type tumors. Data are presented as means ± SD (error bars represent SD). For each group, n = 8 mice, each mouse contributed one data point. Statistical significance was assessed by one-way analysis of variance with Tukey post hoc test. ** = P < .01, *** = P < .001. a.u. = arbitrary units, ns = not significant.

The classification of tumors into angiogenesis-type and vessel co-option–type based on PAM-derived vascular features was further validated by a blinded analysis, as described in Appendix S2.

Treatment of CRLMs

On validation of successful model establishment, we selected two groups from each of the HT29-luc and HCT116-luc tumor-bearing mice for treatment. Bevacizumab (Avastin; Roche Pharma) was administered via intraperitoneal injection at a dose of 10 mg/kg body weight twice weekly for the durations specified in the Table.

Histologic and Immunofluorescence Analysis

Following PAM imaging, the mice were immediately euthanized by cervical dislocation under deep isoflurane anesthesia, and their liver tissues were collected for subsequent histologic analysis. The tissue slices were processed for hematoxylin-eosin and immunofluorescence staining with antibodies, including anti-CD31 and anti-CD105, to identify the blood supply patterns of tumors. The HGPs of lesions were evaluated using ImageJ software (23). Image analysis was performed with either an Olympus BX53 inverted epifluorescence microscope or a Zeiss LSM 800 confocal microscope. Detailed protocols and a complete list of reagents are provided in Appendix S3 and Table S1, respectively.

Statistical Analysis

Statistical analysis was performed with Origin software (version 2025b, OriginLab; RRID:SCR_014212) and GraphPad Prism (version 10.5.0, GraphPad Software; RRID:SCR_002798) by two authors (K.W. and M.W.). The Shapiro-Wilk test was used to assess the normality of continuous variables. Normally distributed data were expressed as means ± SD. For comparisons between two groups, an independent-samples t test was used. For comparisons among multiple groups, one-way analysis of variance was applied; if the overall analysis of variance was statistically significant, post hoc pairwise comparisons were performed using the Tukey test (for comparisons among all groups) or the Dunnett test (for comparisons against a single control group). In cases where the assumption of homogeneity of variances was violated, the Welch test was used, followed by the Dunnett T3 post hoc test. Sample size was determined based on previous similar studies and our pilot experiments, which indicated that the current number of samples per group was sufficient to detect meaningful differences. P values less than .05 (two-tailed) were considered indicative of a statistically significant difference between groups.

Results

Discrimination between Angiogenesis and Vessel Co-Option Subtypes in the Blood Supply Patterns of CRLMs through High-Resolution PAM

A schematic of the PAM method for CRLM imaging is shown in Figure 1, along with an outline of the experimental system setup. After detecting tumor fluorescence with an in vivo imaging system, we exposed the livers for morphologic assessment and then performed PAM scanning. In macroscopic pathology examination of the exposed livers, we found no apparent between-subtype differences in nodule size, color, texture, or distribution in the model mice at the early stage. We scanned lesion areas in both angiogenesis-type and vessel co-option–type tumor models using the PAM system shown in Figure 1. Analysis of high-resolution PAM images revealed distinct vascular architectures at the tumor–liver interface. Angiogenesis-type tumors displayed the characteristic pushing margin, with photoacoustic signals forming an arc-like boundary, whereas vessel co-option–type tumors showed seamless integration of host and tumor vasculature without displacement (Fig 2A–2C).

Angiogenesis-type tumors exhibited lower vessel density but larger vessel diameter than vessel co-option tumors (vessel density: control, 69.34% ± 2.95; angiogenesis-type, 15.66% ± 4.38; vessel co-option–type, 31.44% ± 6.95 [P < .001]; vessel diameter: control, 12.08 μm ± 1.61; angiogenesis-type, 31.45 μm ± 4.59; vessel co-option–type, 13.68 μm ± 1.61 [P < .001]) (Fig 2D–2E). Lacunarity analysis revealed greater structural irregularities and heterogeneity in angiogenesis-type tumors than vessel co-option subtypes (control: 0.014 arbitrary units [au] ± 0.002; angiogenesis-type: 0.294 au ± 0.095; vessel co-option–type: 0.148 au ± 0.059 [P < .001]) (Fig 2F). Analysis of vessel junction density confirmed sparser vascular connectivity in angiogenesis-type tumors (control, 6.53 × 10^-3 n/px² ± 0.41; angiogenesis-type, 1.19 × 10^-3 n/px² ± 0.52; vessel co-option–type, 3.12 × 10^-3 n/px² ± 1.11 [P < .001]) (Fig 2G). To objectively validate the diagnostic accuracy of this PAM method, a blinded assessment was conducted. The validation analysis revealed that the overall classification accuracy based on the aforementioned vascular features reached 87.5% (28 correct of 32 cases; Table S2). This independent validation confirmed that the vascular patterns identified by PAM can serve as objective and reproducible biomarkers capable of accurately distinguishing between different vascular subtypes of CRLM without prior knowledge.

Discrimination between Angiogenesis and Vessel Co-Option Subtypes at Different Stages of CRLM Using High-Resolution PAM

Given that single-time-point PAM imaging may fail to characterize dynamic transitions between two blood supply patterns during tumor progression, we implemented an 8-week longitudinal monitoring protocol (Fig 3A). Dividing the observation period into early and advanced stages revealed distinct blood supply patterns between the tumor subtypes. Gross pathology analysis of the liver revealed distinct morphologic patterns between angiogenesis-type and vessel co-option–type metastatic tumors. Angiogenesis-type tumors progressively expanded, often forming white nodular protrusions that penetrated the hepatic capsule. These enlarging lesions showed limited multifocality, typically clustering in localized regions. Vessel co-option–type tumors, however, appeared as numerous small, diffuse lesions scattered throughout the hepatic parenchyma. Although these vessel co-option metastatic lesions maintained relatively small volumes (angiogenesis-type: 22.78 mm³ ± 14.51 vs vessel co-option–type: 0.96 mm³ ± 0.62 [P < .001]) (Fig S1A), the total number of metastases in the liver per animal was higher than that for angiogenesis-type tumors (angiogenesis-type: 3 ± 0.76 vs vessel co-option–type: 8.5 ± 3.0 [P = .001]) (Fig 3B).

Figure 3:

Photoacoustic microscopy (PAM) imaging and quantitative analyses of microvascular features across tumor subtypes at early and advanced disease stages. (A) Experimental timeline. BLI = bioluminescence imaging. (B) Gross morphologic characteristics of angiogenesis-type and vessel co-option tumor specimens. The specific analysis sites within the multifocal tumor images are marked with asterisks (*). (C) PAM scans of early- and advanced-stage angiogenesis-type colorectal liver metastasis (CRLM) alongside corresponding vessel co-option–type CRLM. (D) Magnified detailed views of PAM scans for both tumor types at different developmental stages. (E) Binarized versions of these PAM scans. (F–I) Quantitative comparison of (F) vessel density, (G) vessel diameter, (H) lacunarity, and (I) vessel junction density between angiogenesis-type and vessel co-option–type CRLMs at early and advanced disease stages. Data are presented as means ± SD (error bars represent SD). For each tumor subtype and time point, n = 8 mice, each mouse contributed one data point. Statistical significance was assessed by independent-samples t test. ** = P < .01, *** = P < .001. Panel A was created with BioRender.com.

Using PAI, we demonstrated persistent differences in vascular parameters (vessel density, vessel diameter, lacunarity, vessel junction density) between angiogenesis-type and vessel co-option–type tumors, for which we observed no convergence during progression (Fig 3C–3I). Early-stage angiogenesis-dependent tumors exhibited lower vessel density (angiogenesis-type: 15.66% ± 4.38 vs vessel co-option–type: 31.44% ± 6.95 [P < .001]) and larger vessel diameters (angiogenesis-type: 31.45 μm ± 4.59 vs vessel co-option–type: 13.68 μm ± 1.61 [P < .001]). These patterns persisted in advanced-stage measurements: Vessel density remained reduced in angiogenesis-dependent tumors (angiogenesis-type: 13.74% ± 4.91 vs vessel co-option–type: 27.99% ± 3.03 [P < .001]), while vessel diameters were consistently larger (angiogenesis-type: 37.69 μm ± 1.49 vs vessel co-option–type: 14.71 μm ± 0.99 [P < .001]). Furthermore, the lacunarity remained greater in angiogenesis-type tumors at both the early (0.294 au ± 0.095 vs 0.148 au ± 0.059 [P = .003]) and advanced stages (0.364 au ± 0.178 vs 0.144 au ± 0.064 [P = .010]). Vessel junction density was persistently lower in angiogenesis-dependent tumors across stages (early stage: angiogenesis-type: 1.19 × 10^-3 n/px² ± 0.52 vs vessel co-option–type: 3.12 × 10^-3 n/px² ± 1.11 [P = .001]; advanced stage: angiogenesis-type: 3.93 × 10^-4 n/px² ± 2.67 vs vessel co-option–type: 2.60 × 10^-3 n/px² ± 0.61 [P < .001]). These findings demonstrate that the distinct blood supply patterns intrinsic to each tumor subtype persist unaltered throughout disease progression.

Evaluation of the Responsiveness of the Two Tumor Subtypes to Bevacizumab via High-Resolution PAM

First, histologic evaluation of pretreatment tumor samples was performed to confirm the baseline experimental status: Neither tumor subtype exhibited significant spontaneous necrosis (Fig S2). Vascular responses in angiogenesis-type tumors following bevacizumab administration were revealed through PAM and validated by histologic analysis (Fig 4A, 4B). After 1 week, quantitative PAM measurements demonstrated comparable vessel density (untreated, 15.66% ± 4.38 vs treated at 1 week, 15.52% ± 3.66 [P = .99]) and vessel junction density (untreated, 1.19 × 10^-3 n/px² ± 0.52 vs treated at 1 week, 7.76 × 10⁻⁴ n/px² ± 4.09 [P = .10]). By 3 weeks, substantial hemodynamic remodeling occurred in treated samples, with PAI showing a 33.3% decrease in vessel density (untreated, 15.66% ± 4.38 vs treated at 3 weeks, 10.44% ± 4.91 [P = .047]). Vessel junction density also decreased (untreated, 1.19 × 10^-3 n/px² ± 0.52 vs treated at 3 weeks, 4.58 × 10⁻⁴ n/px² ± 2.72 [P = .004]) (Fig 4C, 4D). Histologic validation further supported these findings. Specifically, angiogenesis-type tumors exhibited characteristic changes following treatment such as extensive intratumoral necrosis, which corroborates their sensitivity to bevacizumab therapy (Fig 4B). More important, the positive area for CD105, a specific marker for neovessels, was also reduced (untreated, 0.27% ± 0.08 vs treated at 3 weeks, 0.18% ± 0.04 [P = .02]) (Fig S3), providing molecular-level evidence for the effective inhibition of angiogenesis in this subtype by bevacizumab.

Figure 4:

Therapeutic effects of bevacizumab (bev) on distinct tumor subtypes. (A) Photoacoustic microscopy scans of angiogenesis-type and vessel co-option–type tumors following 1 week (1 w) and 3 weeks (3w) of bevacizumab treatment. (B) Corresponding hematoxylin-eosin–stained sections for both tumor subtypes at these time points (10 × magnification; scale bar = 100 μm). Asterisks (*) highlight areas of necrosis. (C) Quantitative comparison of vessel density between distinct tumor subtypes after treatment. (D) Quantitative comparison of vessel junction density between distinct tumor subtypes after treatment. Data are presented as means ± SD (error bars represent SD). For each treatment group and time point, n = 8 mice, each mouse contributed one data point. Statistical significance was assessed by one-way analysis of variance with Dunnett multiple comparisons test. * = P < .05, ** = P < .01. ns = not significant.

Vessel co-option–type tumors exhibited persistent PAM evidence of drug resistance during treatment. We found no evidence of a difference in vessel density (untreated, 31.44% ± 6.95; treated at 1 week, 30.24% ± 6.05; treated at 3 weeks, 28.82% ± 3.56) or vessel junction density (untreated, 3.12 × 10⁻³ n/px²± 1.11; treated at 1 week, 29.5 × 10⁻⁴ n/px²± 5.87; treated at 3 weeks, 27.8 × 10⁻⁴ n/px² ± 5.09) in PAM imaging. Histologic examination further confirmed that vessel co-option–type tumors retained their characteristic architecture after treatment and CD105-positive signals remained consistently low before and after treatment (Fig S3), further confirming that their growth is independent of the neovascularization process targeted by bevacizumab.

The two subtypes exhibited markedly different responses in terms of tumor burden to treatment. The mean volume of angiogenesis-type tumors increased 3 weeks after being subjected to bevacizumab treatment compared with the pretreatment baseline (untreated, 1.25 mm³ ± 0.66 vs treated at 3 weeks, 30.17 mm³ ± 15.55; P = .003). In contrast, we found no evidence of a change in the volume of vessel co-option–type tumors throughout the treatment period (untreated, 0.76 mm³ ± 0.43 vs treated at 3 weeks, 0.81 mm³ ± 0.47; P = .98) (Fig S1B). These findings further confirm, at the macroscopic tumor burden level, the differential responsiveness of these distinct tumor types to the drug.

Histopathologic Validation of Vascular Heterogeneity Based on PAM

The biologic basis of imaging-derived vascular characteristics was confirmed through histopathologic analysis (Fig 5). Hematoxylin-eosin staining (Fig 5A, 5B) and immunofluorescence (Fig 5C, 5D) distinguished two structural patterns: The angiogenesis subtype showed tumor cells with clear boundaries compressing hepatic sinusoids at the tumor–liver interface, altering their native architecture (hematoxylin-eosin, see Fig 5A; immunofluorescence, see Fig 5C). In contrast, the vessel co-option type displayed irregular tumor margins, with tumor cells infiltrating and hijacking hepatocytes rather than forming the pushing border characteristic of the angiogenesis type (hematoxylin-eosin, see Fig 5B; immunofluorescence, see Fig 5D).

Figure 5:

Histopathologic analysis of angiogenesis-type and vessel co-option–type tumors at early and advanced stages. (A) Hematoxylin-eosin (H&E)–stained sections of angiogenesis-type tumors at early and advanced stages. Arrows indicate the tumor–liver interface where tumor cells have clear boundaries compressing hepatic sinusoids, altering their native architecture (10× magnification; scale bar = 100 μm). (B) H&E-stained sections of vessel co-option–type tumors at early and advanced stages. Black circles indicate irregular tumor margins where tumor cells infiltrate and hijack hepatocytes, rather than forming a pushing border (10× magnification; scale bar = 100 μm). (C) Immunofluorescence-stained sections of angiogenesis-type tumors at both stages. Sections were co-stained with anti-CD31 (blood vessels, pink), antihuman serum albumin (HAS) (hepatocytes, red), and antiepithelial cell adhesion molecule (EPCAM) (HCT116-luc cancer cells, green). Nuclei were counterstained with 4’6-diamidino-2-phenylindole (DAPI) (blue) (20× magnification; scale bar = 100 μm). (D) Immunofluorescence-stained sections of vessel co-option–type tumors at both stages. Sections were co-stained with anti-CD31 (blood vessels, pink), anti-HSA (hepatocytes, red), and anti-CK20 (HT29-luc cancer cells, green). Nuclei were counterstained with DAPI (blue) (20× magnification; scale bar = 100 μm). (E) Quantitative analysis of CD31-positive area in immunofluorescence-stained sections from both tumor subtypes at the advanced stage. (F) Immunofluorescence-stained sections of angiogenesis-type tumors at both stages. Sections were co-stained with anti-CD105 (immature blood vessels, red) and anti-EPCAM (HCT116-luc cancer cells, green). Nuclei were counterstained with DAPI (blue) (20× magnification; scale bar = 100 μm). (G) Immunofluorescence-stained sections of vessel co-option–type tumors at both stages. Sections were co-stained with anti-CD105 (immature blood vessels, red), and anti-CK20 (HT29-luc cancer cells, green). Nuclei were counterstained with DAPI (blue) (20× magnification; scale bar = 100 μm). (H) Quantitative analysis of CD105-positive area in immunofluorescence-stained sections from both tumor subtypes at the advanced stage. Data are presented as means ± SD (error bars represent SD). For panels E and H, n = 8 mice, each mouse contributed one data point. Statistical significance was assessed by independent-samples t test. *** = P < .001.

Consistent with the vessel phenotypes identified by PAM imaging, CD31 immunofluorescence revealed a higher microvascular density (or MVD) in the vessel co-option type than the angiogenesis type (CD31 area: angiogenesis-type, 0.36% ± 0.12; vessel co-option–type, 2.57% ± 0.77 [P < .001]) (Fig 5E).

To further distinguish active angiogenesis from a preexisting vessel co-option type, we performed immunofluorescence analysis of CD105 expression, a specific endothelial marker for angiogenesis. The results were markedly different from the overall vascular density pattern revealed by CD31. We observed a CD105 signal in the angiogenesis subtype (Fig 5F), whereas this signal was scarcely detectable in the vessel co-option subtype (Fig 5G). Quantitative analysis demonstrated that the percentage of CD105-positive area was higher in the angiogenesis subtype than the vessel co-option subtype (CD105 area: angiogenesis-type, 0.27% ± 0.08; vessel co-option–type, 0.11% ± 0.03 [P < .001]) (Fig 5H). These pathology findings quantitatively support the vascular parameters derived from PAM and confirm that an active angiogenic process is present in the angiogenesis type, whereas the vessel co-option type primarily relies on use of the host’s preexisting mature vascular network.

Discussion

In this study, we used high-resolution PAM to characterize the distinct microvascular architectures of CRLM in vivo. Using established murine models of angiogenesis-type and vessel co-option–type CRLM, we aimed to directly visualize and quantify vascular features that distinguish these two blood supply patterns. Our quantitative analysis revealed that angiogenesis-type CRLMs exhibited significantly lower vessel density (15.66% ± 4.38 vs 31.44% ± 6.95; P < .001) and lower vessel junction density (1.19 × 10^-3 n/px² ± 0.52 vs 3.12 × 10^-3 n/px² ± 1.11; P = .001) compared with vessel co-option–type CRLMs. Conversely, angiogenesis-type tumors demonstrated higher mean vessel diameter (31.45 µm ± 4.59 vs 13.68 µm ± 1.61; P < .001) and lacunarity (0.294 au ± 0.095 vs 0.148 au ± 0.059; P = .003), reflecting the disorganized and structurally irregular nature of their neovasculature. These microvascular differences persisted throughout the 8-week progression period. Furthermore, longitudinal PAM monitoring revealed a 33.3% reduction in vessel density following 3 weeks of bevacizumab treatment exclusively in angiogenesis-type CRLM (untreated, 15.66% ± 4.38 vs treated, 10.44% ± 4.91; P = .047), with no evidence of a treatment effect in vessel co-option–type tumors (untreated, 31.44% ± 6.95 vs treated, 28.82% ± 3.56; P = .56). To our knowledge, this is the first in vivo experimental evidence that PAM-derived microvascular parameters can reliably discriminate between these distinct blood supply patterns of CRLM and identify vessel co-option–type metastases as resistant to bevacizumab.

Given the distinct HGPs of angiogenesis-type and vessel co-option–type CRLMs, researchers have primarily focused on analyzing HGP composition in prior classification studies (5,24,25). Most study findings suggest that the presence of the replacement HGP is associated with worse survival outcomes and attenuated response to bevacizumab treatment (23–25), a phenomenon attributable to differences in blood supply mechanisms between tumor subtypes. Furthermore, prior research has confirmed that high MVD correlates with poor prognosis of CRLM (26,27). Notably, among distinct metastatic growth patterns, the replacement HGP subtype exhibits significantly higher MVD levels. However, study findings indicate that high MVD alone cannot reliably identify specific CRLM vascular subtypes (8).

In contrast to a previous study (25) in which the authors relied primarily on HGPs to distinguish CRLM subtypes, in our study, we aimed to directly visualize and quantitatively characterize the vascular features. To this end, we used the classic liver metastasis model established using HT29 and HCT116 cell lines, which has been validated in multiple studies to stably recapitulate two distinct types of liver metastases characterized by vessel co-option and active angiogenesis, respectively (10,11). Building on this model, we established a multiparameter evaluation system by integrating vascular morphologic characteristics at the tumor–liver interface with intratumoral vessel density, vessel diameter, lacunarity, and vessel junction density. We found evidence of quantitative differences among morphologically similar tumors. The results demonstrated that angiogenesis gives rise to disorganized neovessels exhibiting excessive structural irregularities and disease-related luminal expansion. Vascular co-option, however, maintained an organized sinusoidal network characterized by a dense branching pattern, near-physiologic vascular diameters, and low geometric irregularity. Contrary to conventional understanding, our quantitative analyses revealed that vessel density and vessel junction density in the co-option subtype were higher than those in angiogenesis-type tumors yet lower than those in normal hepatic tissue. This “intermediate phenotype” provides direct morphologic evidence that cancer cells hijack the host’s preexisting blood vessels to support tumor growth and tissue invasion—a finding consistent with the vessel co-option interpretation proposed by Winkler et al (28) and Donnem et al (29). Functionally, this PAM-based analytical approach complements existing clinical diagnostics by resolving microvascular features that cannot be determined by conventional imaging modalities such as MRI, CT, and US. While the spatial resolution of these conventional clinical techniques is insufficient for quantifying morphologic details, the PAM technique we used, with its 5-µm lateral resolution, enables their quantitative assessment in vivo, potentially informing treatment decisions (20). Crucially, longitudinal PAM monitoring across multiple time points revealed persistent discriminative features throughout extended observation periods.

Within the bevacizumab-treated group, differential responses were observed with PAM imaging. Vessel co-option–type CRLMs exhibited only modest changes in vascular parameters following treatment, whereas angiogenesis-type metastases displayed pronounced vascular regression. In angiogenesis-type tumors, a significant decrease in vessel density was observed after 3 weeks of treatment, consistent with the well-documented penetration barrier for large monoclonal antibodies in solid tumors (30,31). Bevacizumab (molecular weight, 149 kDa) must extravasate and diffuse through the tumor interstitium to neutralize perivascular vascular endothelial growth factor. In angiogenesis-type tumors with aberrant, poorly perfused vasculature and dense stroma, initial drug penetration is inefficient, explaining the preserved vascular density at 1 week. Posttreatment examination confirmed extensive necrosis in angiogenesis-type tumors, whereas the tissue architecture of vessel co-option tumors remained intact, without evident necrotic changes. Notably, although significant vascular regression and extensive necrosis occurred within angiogenic tumors, their total volume increased after treatment. This finding aligns with the “cytostatic” rather than “cytotoxic” nature of antiangiogenic agents (32). In contrast, vessel co-option tumors exhibited a stable volume throughout the course, confirming their nonreliance on the neovascularization targeted by bevacizumab. These findings underscore the fundamental distinction between the two blood supply patterns and support the potential of PAM as a longitudinal therapeutic monitoring method capable of providing reproducible functional evaluation of treatment efficacy.

Beyond guiding systemic therapy, preoperative or intraoperative assessment of blood supply patterns with PAM may inform surgical decision-making. Given the infiltrative growth of vessel co-option–type tumors along preexisting vasculature, complete resection is thought to require wider surgical margins (33,34). Intraoperative identification of a vessel co-option pattern could therefore prompt consideration of extended resection to reduce local recurrence risk. Future development of real-time intraoperative PAM may complement existing navigation modalities for optimized margin control, pending further technical and clinical validation. The vascular phenotype identified by PAM may also have implications for transcatheter arterial chemoembolization. Liver tumors derive their blood supply primarily from the hepatic artery, whereas normal parenchyma is predominantly supplied by the portal vein. Transcatheter arterial chemoembolization exploits this differential by selectively embolizing the arterial supply to tumors. Vessel co-option–type metastases, which PAM can distinguish by their preserved sinusoidal architecture and lack of arterial neovascularization, grow along preexisting sinusoids rather than relying on new vessel formation. Consequently, these tumors may be inherently less susceptible to arterial embolization. We hypothesize that this vascular phenotype could contribute to tumor recurrence following local-regional therapy.

Our study has several limitations. First, the murine model relies on immortalized monoclonal cell lines, which lack the genetic diversity and tumor microenvironment complexity of clinical primary tumors, potentially oversimplifying vascular remodeling mechanisms. Murine models fail to fully replicate the spatiotemporal progression of human liver metastases, possibly compromising the accuracy of antiangiogenic therapy predictions. These constraints mirror a report on the limitations of cell line–derived xenograft models in modeling tumor evolution and heterogeneous treatment responses (35). Second, while our abdominal microincisions are less invasive than conventional liver biopsies, they are still associated with tissue damage risks. Finally, although current PAM systems provide micron-level vascular resolution, their imaging depth remains inadequate for deep hepatic vascular lesions.

In conclusion, high-resolution PAM enables quantitative discrimination of angiogenesis-type and vessel co-option–type CRLMs based on distinct microvascular signatures, and longitudinal monitoring revealed differential responses to bevacizumab aligned with vascular phenotype. These findings support PAM as a functional imaging tool for guiding personalized therapy in CRLM. Future work should validate these results in immunocompetent models and advance miniaturized endoscopic probes and deep-tissue PAM techniques toward minimally invasive clinical translation.

Disclosures of conflicts of interest

Please see ICMJE form(s) for author conflicts of interest. These have been provided as supplemental materials.