Vascular Patterning as Integrative Readout of Complex Molecular and Physiological Signaling by VESsel GENeration Analysis

Abstract

The molecular signaling cascades that regulate angiogenesis and microvascular remodeling are fundamental to normal development, healthy physiology, and pathologies such as inflammation and cancer. Yet quantifying such complex, fractally branching vascular patterns remains difficult. We review application of NASA’s globally available, freely downloadable VESsel GENeration (VESGEN) Analysis software to numerous examples of 2D vascular trees, networks, and tree-network composites. Upon input of a binary vascular image, automated output includes informative vascular maps and quantification of parameters such as tortuosity, fractal dimension, vessel diameter, area, length, number, and branch point. Previous research has demonstrated that cytokines and therapeutics such as vascular endothelial growth factor, basic fibroblast growth factor (fibroblast growth factor-2), transforming growth factor-beta-1, and steroid triamcinolone acetonide specify unique “fingerprint” or “biomarker” vascular patterns that integrate dominant signaling with physiological response. In vivo experimental examples described here include vascular response to keratinocyte growth factor, a novel vessel tortuosity factor; angiogenic inhibition in humanized tumor xenografts by the anti-angiogenesis drug leronlimab; intestinal vascular inflammation with probiotic protection by Saccharomyces boulardii, and a workflow programming of vascular architecture for 3D bioprinting of regenerative tissues from 2D images. Microvascular remodeling in the human retina is described for astronaut risks in microgravity, vessel tortuosity in diabetic retinopathy, and venous occlusive disease.

Introduction

Angiogenesis, lymphangiogenesis, vessel co-option and other microvascular remodeling are required for the pathophysiology of diseases such as cancer, diabetes, and coronary vessel disease, and for normal physiology that includes reproduction, embryonic development, and wound-healing [1–7]. However, the associated vascular patterns are difficult to characterize because of the fractal-based complexity of vascular trees and capillary networks. The VESsel GENeration Analysis (VESGEN) software was developed to automatically map and quantify such vascular branching with parameters that include the fractal dimension, vessel diameter and tortuosity, and densities of vessel length, branchpoint and endpoint. Vascular morphology can be classified into three basic geometric patterns: (1) asymmetric, heterogeneous trees composed of vessels that branch and taper; (2) relatively homogeneous, symmetric networks (plexuses), and (3) tree-network composites [8]. Vascular trees typically develop from immature, capillary-like vasculogenic networks. Capillary networks, are necessarily continuous with the arterial and venous trees of mature tissues and organs.

The VESGEN analysis is based on physiological rules of vertebrate vascular branching that include vessel tapering, bifurcational branching, and the fluid continuum requirements of laminar blood flow. Arterial and venous trees are decomposed by VESGEN into branching generations for site-specific quantification of vascular change, because the functions of small blood vessels, for example, are very different from those of large vessels. For capillary and vasculogenic networks, basic geometric relationships between vessel morphology and avascular spaces (AVSs) are quantified.

To date, the angiogenic stimulators and inhibitors analyzed by VESGEN, including vascular endothelial growth factor (VEGF), basic fibroblast growth factor (bFGF or FGF-2), transforming growth factor-β1 (TGF-β1), angiostatin, keratinocyte growth factor (KGF or fibroblast growth factor-7, FGF-7), and drugs such as leronlimab and the steroid triamcinolone acetonide, have induced a unique ‘fingerprint’ or ‘signature’ vascular pattern that includes cytokine-specific inhibition [9–15,8]. These experimental findings suggest the hypothesis that vascular patterning represents an integrative biomarker or read-out of complex molecular and physiological multi-scale signaling.

In this review, the analysis of vascular trees and networks by VESGEN is illustrated by: (1) retinal changes in crew members after long-duration missions to the International Space Station (ISS); (2) vessel tortuosity during early-stage progression of diabetic retinopathy (DR); (3) central retinal vein occlusion (CRVO) in the human retina; (4) inhibition of angiogenesis in humanized mouse tumors; (5) the unique signature of the epithelial activator KGF as a novel vessel tortuosity factor in an avian model, and (6) the probiotic control of gastrointestinal inflammation in a mouse model. A prototype for the 3D programming of vascular architecture from 2D images of the rat retina for bioprinting of regenerative tissues is described. How the analysis could be adapted to other fractal vascular branching systems such as vein mutations in the wings of Drosophila melanogaster, developing leaf venation in Arabidopsis thaliana [16,17] and neuronal branching is discussed.

VESGEN 2D Software

Description.

VESGEN (version 1.10) is a vascular analysis software developed by the U.S. National Aeronautics and Space Administration (NASA) that is globally and freely available upon request [18]. The software provides an interactive menu-driven user interface (UI) containing three analysis options: Vascular Tree, Vascular Network, and Tree-Network Composite. Detailed explanations of the software are available in the User Guide and elsewhere [8]. Written in Java, VESGEN operates as a sophisticated plug-in bundled with the software ImageJ v1.51r [19]. Image processing algorithms include those from the standard ImageJ toolset and extensive customized analysis (Supplemental Data). Early results for VESGEN were generated by a semi-automated software prototype [9,11,10,14]. The software has been validated by NASA for the following Operating Systems: Microsoft Windows and MacOS (through Mojave; Catalina is currently being tested). A new capability for the automated binarization (segmentation) of grayscale vascular images for subsequent VESGEN analysis by artificial intelligence (AI)/machine learning methods will be included in v1.11, scheduled for release in the near future.

Capabilities.

A user-provided vascular binary image serves as the single input for automated software analysis. User-generated or customized region-of-interest (ROI) options are available. The output consists of specialized vascular maps, spreadsheets of numerical results, and a log documenting the analysis. Vascular output parameters include vessel diameter (D_v), vessel tortuosity (T_v), vessel area density (A_v), vessel length density (L_v), vessel number density (N_v), vessel branch point density (Br_v), end point density (E_v), and the fractal dimension (D_f, measured by a box-counting algorithm). The space-filling properties of complex objects such as vertebrate vascular branching, arboreal trees, coastlines, and angiosperm leaves are measured by fractal mathematics, a non-Euclidean geometry developed by B. Mandelbrot and others [20,21]. These objects are characterized by the geometric property of self-similarity (i.e., repetition of a pattern such as vascular bifurcational branching at increasingly smaller length scales). For 2D images, D_f varies as a fractional value between 1 and 2, the limiting Euclidean dimensions of a line and rectangle. As other examples, L_v is defined as the total vessel length divided by the area of the image or ROI. Vessel tortuosity, T_v, is calculated as the ratio of the arc length of the vessel to the chord length (i.e., distance of the straight line connecting the two end points of the vessel) [22–24]. A minimum value for T_v is therefore 1.0. Parameters are reported in pixel units and additionally in physical units if a microscope calibration factor is entered. To optimize the results, a user can interact with the interface to select other helpful features such as re-combining the branching generations into groups of small, medium and large vessels. Mapping and quantification is based on algorithms for the vessel skeleton and vessel area, and fractal box-counting [9,8]. The algorithms were validated by pixel counting of the skeleton and area vessel density in ImageJ, and by comparing the fractal box-counting algorithm in ImageJ with another box-counting program written in MATLAB^® (The MathWorks, Inc., Natick, MA, USA).

Vascular Trees.

Vascular trees are highly branching, asymmetric structures characterized by tapering vessels. For analysis of successive branching generations by the Vascular Tree option, vessels are assigned to the proper generation by specialized algorithms according to a repertoire of vascular branching rules (Fig. 1) [8]. The averaging of many parameters within a vascular tree such as vessel length or diameter is not generally meaningful because the parameters can vary so greatly over many successive branching generations. Vessels throughout the entire tree image are decomposed by the analysis into successively smaller branching generations (G₁, G₂, … G_x) for which the largest vessel is designated G₁. Vessel assignments are determined by a repertoire of vascular branching rules such as vessel tapering to accommodate a range of vessel sizes within a given generation, given the variability of biological vascular branching. A 1/√2 (0.71) bifurcational condition was chosen to maintain an equal cross-sectional area of two symmetric (dichotomous) offspring vessels compared to the cross-sectional area of a parent vessel, using a user-adjustable default tolerance factor of 15%. As examples, L_v1–3 denotes vessel length L_v with respect to larger branching generations G₁–G₃, and N_v6 is the vessel number density N_v for the smaller branching generation G₆. Nonetheless, the most numerous branching event in most vascular trees is the offshoot of a much smaller vessel from a larger vessel, such as a G₅ vessel from a G₂ vessel [10,11,14,15,8,25,26]. Physical dimensions are notated as in the following example from Figure 1, where L_v (μm/μm²) = 0.00121×10⁻⁴ is denoted as 12.1×10⁻⁴.

Fig. 1. Mapping and quantification by VESGEN of arterial and venous trees.

The vascular tree analysis is illustrated by an image of the left retina of an astronaut crew member acquired after a 6-month mission to the ISS. Binary arterial and venous trees extracted from the grayscale image (a–c) served as sole inputs to the software for the automated generation of arterial and venous maps (d–i). The software first calculates a skeleton (centerline) of the vascular branching that supports calculation of the fractal dimension (D_f) and overall vessel length density (L_v). Together the vascular binary image and skeleton then generate additional mappings such as the distance map and branching generations for calculation of site-specific information on differences in vessel parameters that include small and large vessel diameter, density, and tortuosity. As shown by the legend, colors within the distance map correspond to pixel distance to the edge of the vessel. Arterial and venous D_f were 1.354 and 1.329 and L_v, 12.1×10⁻⁴ and 10.7×10⁻⁴ μm/μm². Average diameter ± SD of arteries and veins per branching generation ranged from 35.5 ± 1.8 and 41.1 ± 3.9 μm for G₁ (largest generation) to 13.2 ± 2.1 and 13.0 ± 11.6 μm for G₅ (smallest). Arterial and venous branching generations (F, I) can be grouped by a user option into groups such as large and small vessels (Fig. 2). Scalebar, 200 μm.

VESGEN, VESsel GENeration; ISS, International Space Station; IR, infrared.

Vascular Networks.

Vascular networks are defined geometrically as non-branching, closed tubular structures that are often symmetric [8,27]. A basic constraint of vascular network analysis is that the fractional areas of network vessels and AVSs (lacunae or holes) must sum to 1, when normalized by the total ROI area. The two structure-function extremes are the highly permeable, thick ‘presinusoidal’ capillary vessels with small AVSs in the liver, compared to highly impermeable, thin capillary vessels with large AVSs constituting the blood brain barrier, as described in an interesting study of these geometric relationships [27]. The Vascular Network option generates results that include A_v, L_v, Br_v, and D_f, histogram data of vessel diameters, the number and areas of AVSs, and their geometric relationship with the vascular network.

Vascular Tree-Network Composites.

Vascular tree-network composites combine the asymmetric, tapering branching of trees with the more symmetric geometry of networks. The composite image can represent a normal vascular tree that is continuous with the capillary bed, as is typical of vascular branching in every vertebrate tissue and organ. The composite could also represent either a pathological neovascularization or a transitional state of development from an immature network to a more mature tree [8]. When the Tree-Network Composite option is selected, vessel parameters are generated that include the Vascular Tree and Vascular Network outputs described above.

Extraction (Segmentation) of Binary Vascular Pattern, New AI Methods, and 3D Analysis.

Vascular patterns were extracted from grayscale clinical or microscopic images and converted to binary (black/white) images using Adobe Photoshop^® (San Jose, CA, USA) because of the convenient layering features, described in greater detail previously [15,8,25]. In earlier studies, images were segmented using ImageJ. Binary vascular patterns from the human retina were further separated into arterial and venous trees using basic principles of anatomy and physiology [25]. For example, arterial and venous trees tend to originate as pairs from the optic disc, and arteries are of smaller diameter than veins. Once a vascular tree was identified as arterial or venous, the tree was followed from its origin at the optic disc to its termination at the smallest generations according to vessel connectivity, bifurcational branching, and tapering (morphological characteristics of a mature vascular tree). Vessel interpretation was decided by agreement between two vascular analysts, subject to final decision by the senior analyst.

Methods of vessel extraction by AI/machine learning and other computational advances are being developed by NASA for the automated extraction (segmentation) of binary vascular patterns from grayscale images [28]. The prototype computer code is incorporated into the VESGEN software and is scheduled for release in 2020 (v1.11). Such AI/machine learning models require training/evaluation datasets from which they learn how to perform a given task. Extensive AI studies were performed using a DR dataset, described below, comprising 34 example cases of grayscale clinical retinal images and the binary (black/white) images of vascular pattern extracted by expert vascular analysts. The DR dataset was further augmented by 40 images from a publicly available dataset of the human retina [29]. The AI/machine learning methods are able to generalize and perform reasonably well on previously unseen images. The prototype AI/machine learning code was used to extract vascular patterns from the avian chorioallantoic membrane (CAM) for the KGF study, followed by further refinement with Photoshop^® (Fig. 4). A prototype for 3D bioprinting capabilities with 3D vascular tubular patterns transformed from 2D vascular images is in early-stage development and summarized in the 3D bioprinting section below.

Microvascular Remodeling and Angiogenesis: Background Methods

Vascular Trees

Recent studies for the astronaut retina and DR described below are being reported in detail elsewhere.

Vascular Decrease in Astronaut Retinas after Long-Duration Missions to the ISS.

Following approval of the retrospective study design by the Institutional Review Board (IRB) and Lifetime Surveillance of Astronaut Health (LSAH) at the NASA Johnson Space Center, eight U.S. astronauts were recruited with written informed consent. All astronaut crew members had completed 6-month missions aboard the ISS with pre- and post-flight exams that included ocular coherence tomography (OCT) by Heidelberg Engineering GmbH Spectralis^®, (Heidelberg, Germany) with retinal vascular images centered on the optic disc acquired by 30° near infrared (IR) mode (Figs. 1–2). Gender, age, and mission of the astronauts were not available due to subject privacy concerns.

Fig. 2. Vascular decrease in the retinas of an astronaut after 6 months on the ISS.

Arterial and venous density decreased in the left and right post-flight retinas of an astronaut crew member. Binary arterial and venous trees extracted from grayscale images of the left and right retinas (a–d) were analyzed by VESGEN as vascular skeletons (e–h, m–p) and generational branching grouped into large (G_v1–4, red) and small (G_v5–6, yellow) generations (i–l, q–t). Decreases in vascular parameters were calculated from skeletonized and grouped maps (Table 1). For visual comparison, all images were aligned with optic nerve to left (i.e., images of right retina rotated 180°). Scalebars, 200 μm.

Progression of DR.

Following approval of the University of Florida IRB, 39 diabetic patients at varying stages of DR ranging from mild to moderate or severe nonproliferative DR (NPDR), as well as 39 age- and sex-matched healthy controls, were recruited with written informed consent. Retinas of the subjects were photographed at 12.5 μm/pixel by 30° Heidelberg Spectralis OCT following fluorescein angiography (FA) according to standards of the Early Treatment Diabetic Retinopathy Study (ETDRS). Altogether, 34 images that standardized image resolution and retinal field of view and further displayed sufficient clarity for the vascular analysis were analyzed by VESGEN.

KGF as Novel Vessel Tortuosity Factor from a Developmental Model.

In a series of four experiments, fertilized eggs of Japanese quail (Coturnix coturnix japonica) were cracked at embryonic day three (E3) and cultured further in 6-well Petri dishes in a tissue culture incubator as reported previously [9–12,14,13,15,8] for which the protocol is available [30]. At E7, 0.5 ml phosphate buffered saline (PBS) containing 10 μg of recombinant human KGF, the kind gift of Andrew G. Farr, University of Washington [31], was applied to the CAM surface. The CAMs were aldehyde-fixed and dissected after incubation for 48 hours. Results reported here for two images acquired by brightfield microscopy are representative of vascular response at 2.5–20 μg KGF/CAM following 24 or 48 hours of incubation. The images of end points in the arterial tree were separated from the venous tree and binarized for VESGEN analysis by a well-established artifact of CAM fixation, in which blood is retained primarily within arteries but not within veins.

Inhibition of Tumor Angiogenesis in Humanized Mouse Models.

Twenty-four hours after 2.25 Gy whole body X-ray irradiation, male NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ, commonly known as the NOD scid IL-2 receptor gamma knockout (NSG^™, The Jackson Laboratory, Bar Harbor, ME, USA) mice were engrafted with human bone marrow (BM) cells. De-identified human donor cells were obtained by back-flushing filter packs utilized by the Cleveland Clinic Bone Marrow Transplant program. Fresh, non-frozen leukocytes were purified by Ficoll-Hypaque gradient centrifugation, washed in PBS, and assessed for viability (Vi-Cell^™, Beckman Coulter, Inc., Brea, CA, USA). Human BM mononuclear leukocytes were injected into the lateral tail vein (10⁶ cells/mouse). Twenty-four hours later mice were inoculated intradermally in the flanks with 2×10⁶ SW480 human colon carcinoma cells (ATCC, Manassas, VA, USA). The next day mice were randomized into control and treatment groups of 8 animals each by body weight. Leronlimab (CytoDyn, Inc., Vancouver, WA, USA) or normal human IgG was administered intraperitoneally (i.p.) at 2.0 mg/mouse twice weekly.

Blood vessels growing at the periphery of dermally inoculated day 10 SW480 colon carcinoma tumors in humanized NSG^™ mice were photographed using a dissecting microscope at 12.5× magnification. Two measurements were taken to assess the tumor area (the largest diameter coplanar with the skin, and a second diameter perpendicular to the first). The product of these two measurements was used as an index of tumor area. Images were captured using an operating microscope with 12.5 objective lens (World Precision Instruments, PSMT5, Sarasota, FL, USA). Tumor photographs were subjected to digital analysis using VESGEN software, where the region of interest representing the tumor mass defined the perimeter of the tumor. The output was a series of colored vessel generation maps (colored vessels on black background) in which the vessels of largest diameter were defined as G₁ (red), with each subsequent smaller generation represented as G₂–G₉. From these maps, the software calculated the total vessel area, vessel length density, vessel number, and vessel diameter.

Vascular Networks

Probiotic Protection in a Mouse Model of Intestinal Inflammation.

Unpublished images of the luminal capillary network in the mouse colon acquired for a previous study [32] on gastrointestinal protection against inflammation by the probiotic yeast Saccharomyces boulardii (S. boulardii) were mapped and quantified with the Vascular Network option. In brief, 8-week-old female C57BL6 mice received 4% dextran sulfate sodium (DSS) for 5 days. Administration of S. boulardii was by daily oral gavage at 6×10⁸ CFU. Thereafter, blood vessels were labelled by retro-orbital injection of fluorescein isothiocyanate prior to sacrifice of mice. Colon tissues were immediately removed and imaged by confocal microscopy, from which the 3D image reconstructions acquired to assess blood vessel volume and density were reduced to 2D grayscale images and binarized to black-and-white vascular patterns (n = 5). For the VESGEN analysis, isolated cells and (infrequent) small vessel fragments were deleted, so that only connected vascular networks remained.

Tree-Network Composites

Vascular Changes with Central Retinal Vein Occlusion in the Human Retina.

A normal eye of a healthy 25 year old male served as the normal control for the eye of a patient diagnosed with CRVO and treatment history of multiple intravitreal bevacizumab injections (55 year old male without significant past medical history), and the patient’s fellow unaffected eye. The retinas were imaged by FA with adaptive optics scanning laser ophthalmoscopy (AOSLO) as described previously [33]. Binary (black/white) vascular patterns extracted from the images were mapped by VESGEN with the Tree-Network option.

Vascularized Tissue-Specific 3D Bioprinting from 2D Vascular Patterns.

Briefly, the eye of a 3-month old male Long-Evans rat was enucleated, fixed in 4% paraformaldehyde for 24 hr, stored in PBS, and retinal flat mount dissected and immunolabelled with endothelial cell-specific Isolectin GS-IB₄ biotin conjugate (Invitrogen - I21414, Thermo Fisher Scientific, Inc., Waltham, MA, USA) and Alexa Fluor 488 (Invitrogen S-32354) according to previous methods [34]. Acquired with a Leica SP8 confocal microscope (Leica Microsystems, Wetzlar, Germany) at 20× magnification and 2048×2048 pixel resolution using 488 nm excitation, the vascular image was post-processed into a binary vascular pattern and analyzed with VESGEN using the Vascular Tree-Network Composite option as described above. The prototype for the VESGEN 3D software, based in part on 3D transformations of VESGEN 2D algorithms, is being developed by NASA with additional new 3D algorithms and visualizations. The 3D software capabilities will include the conversion of a vascular 2D image into a 3D tubular vascular system that, as one example, can be used for the ‘blueprinting’ of tissue-specific blood vessels within 3D-bioprinted tissues and organs for regeneration and replacement [35,18]. From a stack of 2D vascular pattern images, a 3D skeleton representing the centerline of the vessels was extracted in ImageJ. A set of branches was then derived from the 3D skeleton using an enhanced version of the AnalyzeSkeleton plugin that outputs the following attributes along the branches: branch point type (e.g., junction, slab, or end points), 3D distance map, and branch generation. To construct a 3D tube for each branch, the following steps were performed: (1) computation of a smooth curve that interpolates the branch for (2) a set of points along the curve, constructing a circular disc, whose diameter is based on the 3D distance map value at that point, and (3) construction of 3D polygons by connecting points along the neighboring discs. After the 3D tube was constructed, the tube was rendered as a 2D image and colored by the corresponding branch generation attribute. The ultimate goal is to create a dataset with the necessary information for vascular 3D bioprinting. A printable .stl file for 3D printing was generated using the Bio-CAD software from the Bioprinting Suite of the 3DDiscovery^™ bioprinter (regenHU Ltd., Switzerland). For the bioink surrogate, a glycerin-based cream (NIVEA Crème, Beiersdorf, Germany) was supplemented with commercial food dyes of red, green, and blue color to match the VESGEN generated image.

Statistical Analysis

Results of the clinical and animal in vivo studies are expressed as either mean ± standard deviation (x ± SD) or as mean ± standard error (x ± SE). Values of comparative groups were calculated using a conventional paired t test, for which a p value of <0.05 was considered significant.

Vascular Trees

Vascular Decrease in the Retinas of Astronauts after Long-Duration Missions to ISS

Recent studies reveal that significant risks for ocular and visual damage are incurred by astronauts exposed to microgravity on the International Space Station (ISS) during long-duration missions [36,37,19,38]. Denoted by NASA as the Spaceflight-Associated Neuro-ocular Syndrome (SANS; formerly Visual Impairment and Intracranial Pressure syndrome, VIIP), symptoms include optic disc edema (ODE), increased retinal and choroidal thickness, cotton wool spots, and decreased near visual acuity. We tested the hypothesis that blood vessels in the human retina remodel in response to the physiological stresses of prolonged microgravity and may provide a sensitive, early-stage predictor of SANS susceptibility. Post-flight arterial and venous decreases were quantified by D_f and L_v in 11/16 retinas (paper in review). Losses resulted primarily from a decrease in density of small vessels. Overall, vascular decreases ranged from low up to highest in the single retina diagnosed with SANS. Relative to the SANS retina, high losses in arterial and venous patterning also occurred bilaterally in another astronaut (Fig. 2, Table 1).

Table 1. Decreased vascular patterning in the retinas of an astronaut after 6-month mission on the ISS.

Arterial and venous densities from the skeletonized (linearized) and grouped generation maps of the left and right retinas of an astronaut (Fig. 2) were quantified by two confirming measures, D_f and parameters for L_v.

Vascular tree	Parameter	Symbol	Left, pre-flight	Left, post-flight	Right, pre-flight	Right, post-flight
Arterial	Fractal dimension	D f	1.355	1.310	1.367	1.316
	Length density
	All vessels	L v	12.33	9.94	12.98	10.16
	Large vessels	L vl-4	7.90	7.71	8.36	7.84
	Small vessels	L v≥5	4.43	2.23	4.62	2.29
Venous	Fractal dimension	D f	1.326	1.284	1.332	1.290
	Length density
	All vessels	L v	10.53	8.64	10.81	9.02
	Large vessels	L v1-4	5.64	5.86	5.08	5.73
	Small vessels	L v≥5	4.89	2.82	5.58	3.47

D_f, fractal dimension (dimensionless); L_v, vessel length density. Physical dimensions of L_v, L_v1–4, L_v≥5, ×10⁻⁴ μm/μm².

Currently it is not known whether substantial post-flight vascular decreases in the astronaut retinas result from vaso-obliteration (vascular dropout or loss) or from decreases in vessel diameter below the limit of image resolution. Studies of human response to microgravity and space radiation health risks are of increasing importance, given the current global interest in lunar colonization and deep space exploration. In general, research with VESGEN has revealed that the smaller vessels remodel most actively during progressive vascular-dependent processes such as DR and molecular regulation of angiogenesis and anti-angiogenesis [11,10,14,15,8,25,39,32,26].

Remodeling of Arterial and Venous Trees with Progression of Diabetic Retinopathy

DR remains the major blinding disease of working-aged adults in industrialized countries [40]. The early stages of mild, moderate, and severe NPDR are defined by a progressive change of vascular status that precedes the pathologically excessive neovascularization of late-stage proliferative DR (PDR). Previous VESGEN research revealed a homeostatic-like alternation or oscillation of retinal vessel density [25], during which vessels increased or proliferated during moderate NPDR compared to mild NPDR, prior to a second vascular dropout during severe NPDR and final neovascularization during PDR. We therefore continue to investigate DR for better understanding of potential approaches to reverse or stop early-stage progression. As demonstrated by various lines of evidence, for example, activation of the vasoprotective arm of the renin-angiotensin system (RAS) within hematopoietic stem/progenitor cells (HSPC) may offer a promising early-stage strategy for prevention or reversal of early-stage progression during NPDR [41–44]. Early results of the current larger DR study show that arterial and venous density, particularly of the small vessels, increased with NPDR progression (Fig. 3, Table 2).

Fig. 3. Remodeling of arterial and venous trees with progression of nonproliferative diabetic retinopathy (NPDR).

Clinical images of the human retina with mild (a, d) and moderate (G) stages of NPDR by the increasing presence of established clinical disease markers such as microaneurysms and hemorrhagic leakage. One image (d) is representative of a subset of retinas within the study (approximately 13%) displaying the additional phenotype of unusually tortuous vessels (Table 2). Arterial (b, e, h) and venous (c, f, i) maps illustrate branching generations from G₁ to G₇ (legend). The generational maps were grouped into large (G_1–4) and small (G_≥5) generations to further quantify site-specific changes within the complex branching trees using the VESGEN generation grouping feature (Table 2). Scalebars, 200 μm.

Table 2. Progression of nonproliferative diabetic retinopathy (NPDR) in arterial and venous trees of the human retina.

Arterial and venous densities are quantified from maps of three human subjects diagnosed with mild to moderate NPDR (Fig. 3) by the fractal dimension (D_f) and vessel length density, L_v. Site-specific vascular differences are further characterized for branching generations one to four (L_v1–4) and generations greater than or equal to five (L_v≥5). Tortuous vessels in one image (Fig. 3d) characteristic of a subset of retinas are compared to other images by a sensitive measure of vessel tortuosity (T_v), geometrically constrained as ≥1.0.

Vascular tree	Parameter	Symbol	Mild NPDR (Fig. 3a)	Mild NPDR (Fig. 3d)	Moderate NPDR (Fig. 3g)
Arterial	Fractal dimension	D f	1.320	1.333	1.377
	Length density
	All vessels	L v	1.571	1.650	2.100
	Large vessels	L vl-4	0.885	0.733	0.854
	Small vessels	L v≥5	0.686	0.916	1.245
	Tortuosity
	All vessels	T v	1.150	1.190	1.130
	Large vessels	T vl-4	1.124	1.149	1.109
	Small vessels	T v≥5	1.186	1.227	1.145
Venous	Fractal dimension	D f	1.363	1.383	1.399
	Length density
	All vessels	L v	1.947	2.129	2.332
	Large vessels	L vl-4	0.577	0.883	0.702
	Small vessels	L v≥5	1.370	1.246	1.633
	Tortuosity
	All vessels	T v	1.153	1.184	1.153
	Large vessels	T vl-4	1.100	1.191	1.138
	Small vessels	T v≥5	1.177	1.180	1.159

Physical dimensions of L_v, L_v1–4, L_v≥5, ×10⁻² μm/μm² and T_v, pixel/pixel (D_f, dimensionless).

Of the images analyzed for this study, approximately 13% display the additional phenotype of pronounced vessel tortuosity, particularly of veins. In this and previous VESGEN studies, the overall vessel tortuosity index (T_v), a sensitive indicator, typically ranges for most vascular retinal images from approximately 1.11 to 1.15. For the tortuous arteries and veins in the NPDR retina (Fig. 3d), T_v is 1.18 and 1.19, respectively. Retinal vascular tortuosity is associated with numerous vascular-dependent diseases such as systemic hypertension, DR, and retinopathy of prematurity (ROP) [23,45]. Development of vascular tortuosity in the retina is believed to result from multiple interactive processes such as angiogenesis, changes in blood flow and blood pressure, genetic factors, and vascular degeneration. Retinal tortuosity may represent or predict similar microvascular changes occurring elsewhere in the body, such as in the brain [24].

Epithelial Activator KGF as Novel Vessel Tortuosity Factor

Produced throughout the body by mesenchymal cells such as fibroblasts, KGF is a paracrine activator of proliferation and migration via FGF receptor isoforms in epithelium, but not endothelium [46]. Described initially as a mitogen for keratinocytes in wound healing, KGF has further demonstrated cytoprotective properties by promotion of intestinal repair [46], suppression of malignant epithelial phenotypes [47], and as a therapy, protection against radiologic, chemotherapeutic, and cytotoxic agents [48]. On the other hand, KGF and its interactions with VEGF and TGF-β2 have been implicated in the progression of choroidal neovascularization (CNV) [49]. A major cause of vision loss, CNV is the growth of new blood vessels from the underlying choroid into the sub-retinal pigment epithelium via a break in Bruch’s membrane. In human pancreatic ductal epithelial cells, activation by KGF of the proinflammatory transcription factor family, nuclear factor kappaB (NF-κB), further stimulated the expression of VEGF and matrix metalloprotease-9 (MMP-9), a protease family involved in the degradation of the extracellular matrix [50].

In a developmental model of angiogenesis in the avian CAM, KGF strongly increased arterial tortuosity, accompanied by some increase in arterial density (Fig. 4, Table 3). To our knowledge, this is the first report of KGF as a regulator of vessel tortuosity. Interestingly, in both the tortuous NPDR retina (Fig. 3d–f) and this KGF-treated developmental model, the quasi-sinusoidal pattern of the larger vessels is repeated by the small vessels at a smaller scale of decreasing amplitude and periodicity. The fractally scaling pattern therefore appears to be a function of vessel diameter and is perhaps determined by the cardiovascular (Womersley) fluid dynamics of pulsatile blood flow. Signaling by epithelial activator KGF may have reduced the stiffness of the chorionic epithelium surrounding the blood vessels, allowing the sinusoidal-like curvature of the vasculature to be determined by the pulsatile flow. Whether KGF is involved in the signaling of the subset of NPDR patients displaying significant vessel tortuosity (Fig. 3) is currently not known.

Fig. 4. Epithelial activator KGF induces vessel tortuosity in vivo.

The quail CAM, avian analog of the placenta, was treated (a) with and (d) without KGF at 10 μg CAM for 48 hr. Images generated by VESGEN of the arterial end points include distance maps (b, e) where legend indicates pixel distance to vessel edge and branching generations with legend (G₁–G₇; c, f) that support quantification of specific generational changes exerted by KGF (Table 3). As a novel regulator of vessel tortuosity, KGF may have increased the activity of matrix metalloproteases in the chorionic epithelium [50], thereby decreasing tissue resistance to sinusoidal vascular patterning by the pulsatile blood flow. Scalebar, 500 μm (d).

Table 3. KGF as regulator of vessel tortuosity.

Vascular patterning in the arterial end points of a quail CAM treated with KGF (10 μg) is compared to untreated control (Fig. 2). Quantification of the skeletonized arterial patterns includes vessel tortuosity (T_v) and two measures of vascular space-filling capacity, the fractal dimension (D_f) and total vessel length density (L_v).

Vascular tree	Parameter	Symbol	Control	KGF, 10 μg/CAM
Arterial	Fractal dimension	D f	1.463	1.506
	Length density
	All vessels	L v	3.866	4.515
	Large vessels	L vl	0.113	0.123
	Medium vessels	L v2–5	2.081	1.718
	Small vessels	L v≥6	1.602	2.621
	Tortuosity
	All vessels	T v	1.150	1.205
	Large vessels	T v1	1.104	1.109
	Medium vessels	T v2–5	1.154	1.288
	Small vessels	T v≥6	1.149	1.161

Physical dimensions of L_v, L_v1–4, L_v≥5, ×10^–2 μm/μm² and T_v, pixel/pixel; D_f, dimensionless.

The CAM assay has been used by us and others as an analog of retinal vascular branching pattern because of its morphological similarity to the retina. We have found the CAM to be a useful, optically accessible model for testing and quantifying vascular responses to numerous cytokines and therapeutics. Because this assay does not visualize venous trees, the effects of KGF and other molecular regulators on veins is not known beyond an observational level.

Inhibition of Tumor Angiogenesis in Humanized Mouse Models

Analysis of blood vessels at the periphery of day 10 SW480 colon carcinoma tumors inoculated in the dermis of the flanks revealed a significant decrease in the number of vessels in tumors from leronlimab-treated humanized mice compared with IgG-treated hosts (Fig. 5), consistent with leronlimab causing an inhibitory effect on neoangiogenesis in the tumor bed. Utilization of VESGEN software allowed detailed comparisons between the two treatment groups and revealed marked reduction in key parameters of the vascular network feeding the tumor, including 62% reduction in total vessel area (pixels) (p = 0.013), 53% reduction in vessel length density (p = 0.0011), 61% reduction in number of large vessels (p = 0.0082) and 80% reduction in number of small vessels (p = 0.017).

Fig. 5. Inhibition of tumor angiogenesis in a humanized mouse model.

Immuno-incompetent mice (NSG^™) were humanized by inoculation with normal human bone marrow derived leukocytes. Then SW480 human colon carcinoma cells were injected intradermally in the mouse flanks. Mice were treated either with normal human IgG (left columns) or leronlimab (right columns) at a dose of 2 mg/kg i.p. twice weekly. On day 10 the mice were euthanized, and the tumor inoculation site was photographed and subjected to VESGEN analysis.

Vascular Networks

Probiotic Protection in a Mouse Model of Intestinal Inflammation

Angiogenesis is required for the complex responses of inflammatory bowel diseases (IBD) and for intestinal mucosal remodeling during recovery [51,52]. In the mouse, DSS leads to a reversible colitis characterized by inflammation and angiogenesis [53,54]. The model was therefore used to study the effects of S. boulardii, a probiotic yeast, on remodeling of the colonic capillary vasculature during colitis induction [32]. The yeast protects against IBD, diarrhea, tumor formation, and intestinal injury by multiple immunological mechanisms [55,56]. Nonetheless, regulation of protective mucosal host responses by S. boulardii is not yet fully understood. The purpose of this previously reported study was to investigate how the probiotic alters signaling by vascular endothelial growth factor receptors (VEGFRs) as fundamental regulators of angiogenesis.

The luminal microvasculature of the normal intestine is organized as a highly regular lattice or network comprised of subepithelial capillaries surrounding the mucosal glands [57]. Closed colonic vessels that appear ‘open’ or blind-ended in 2D images actually turn into the z-plane to connect arterioles and venules with the underlying crypt circulation, resulting in some unconnected areas (Fig. 6) of the otherwise regular lattice spacing of the enclosed AVSs. The colonic capillaries are of small caliber (diameter) relative to the surrounding tissue and therefore, have a small fractional ratio of vessel area compared to the AVS, unlike large sinusoidal capillaries of the liver and lung vascular networks [8,27]. Mapping and quantification of vascular networks by VESGEN excludes any AVSs located at the edge of the image because representation of an AVS is incomplete in those regions and its vascular density, therefore, is unknown.

Fig. 6. Network analysis of pathological angiogenesis with probiotic protection during intestinal inflammation.

(first row) In confocal fluorescent images, the luminal capillary network of the normal mouse colon appears as a regular lattice structure (left column) that is disrupted by inflammation following administration of DSS (middle column). Treatment with S. boulardii, a yeast probiotic (right column), moderated the inflammatory angiogenic response. (second row) Binary vascular patterns extracted from confocal images were mapped and quantified as described previously with the Vascular Network option [32], in which the entire vascular pattern is analyzed for vessel parameters and fractional vascular/avascular areas (Table 4). However, only regions containing complete AVS (black) are quantified for other network parameters. Scalebar, 300 μm.

The highly regular lattice network of luminal capillaries in the normal mouse colon was disrupted by inflammation and angiogenesis following DSS administration (Fig. 6). Compared to control mice, the vascular inflammatory phenotype displayed statistically significant increases in vessel branch points, end points, fractional vascular area, and numbers of AVSs (holes) reported previously (all p < 0.05) [32]. In addition, the mean area per AVS decreased and varied considerably in size as in the representative, previously unpublished images shown here (Fig. 6, Table 4). Although vessel diameter increased with DSS administration, differences were not statistically significant. Treatment with S. boulardii considerably restored network uniformity as demonstrated by reduced variation in the size of the AVS.

Table 4. Inflammation and abnormal angiogenesis in the microvascular network of mouse intestine.

Results were calculated for either the entire image or fully enclosed AVS from Figure 5. Disruption by the DSS model of intestinal inflammation is compared to normal vascular networks and probiotic protection by Saccharomyces boulardii (Sb). Results expressed as mean ± SD according to microscope calibration factor, 1.48 μm/pixel.

Analysis	Groups	Control	DSS	DSS + Sb
Entire image	Vessel diameter, μm	16.6 ± 5.4	17.5 ± 9.5	19.2 ± 8.9
	Number of branch points	304	713	425
	Number of end points	132	432	222
	Fractional vascular area	0.270	0.349	0.349
	Fractional avascular area	0.730	0.651	0.651
Enclosed AVSs	Number of AVS	90	163	115
	Area per AVS, μm2	11,207 ± 7,444	4376 ± 9669	6,395 ± 8,182

To demonstrate association with other inflammatory measures, moderation of the DSS inflammatory angiogenic response by S. boulardii was accompanied by reduced histologic scores of mouse colonic inflammation and weight loss, reduced angiogenesis induced by VEGF in a mouse ear model, and reduced capillary tube formation in an endothelial cell culture model characterized by increased VEGFR-2 phosphorylation and activation of downstream kinases PLCγ and Erk1/2 [32]. In another study of the colonic microvascular network by VESGEN analysis [39], a statistically based linkage analysis of the ‘open’ but regular AVS was developed to further account for uniformity of the normal network compared to network disruption by DSS. In summary, quantification by the Vascular Network option characterized how inflammatory angiogenesis increased vessel density and disrupted regular lattice geometry within the normal colonic microvascular network. Probiotic treatment by S. boulardii significantly restored the vascular network to a more normal morphology by reducing the angiogenic response.

Vascular Tree-Network Composites

Vascular Changes with Central Retinal Vein Occlusion in the Human Retina

As a recent revolution in ophthalmic imaging [33], AOSLO-FA, and other approaches can now visualize most or essentially all of the vessels within the retina (Fig. 7, Table 5). At this level of magnification, the Tree-Network Composite analysis is appropriate because the capillaries are continuous with the pre-capillary arterioles and post-capillary venules. The 2D rendering of these retinal vascular trees that in reality are somewhat 3D does result in some artifactual vessel crossovers, so a vascular branchpoint analysis would not be meaningful.

Fig. 7. Tree-Network Analysis of microvascular changes in the human retina from CRVO.

Vessels in the parafoveal region of the human retina were imaged by AOSLO-FA [33] and analyzed by the VESGEN Tree-Network Composite option. Compared to the control retina of a healthy young adult, vessel density within the CRVO retina of this middle-aged subject was reduced and irregular (Table 5). Vascular damage was accompanied by prominent microaneurysms, considerable fluorescein leakage from the vessels and apparent vascular dropout in the central avascular foveal region. Vessel changes in the fellow retina of the CRVO subject were intermediate between that of the control and CRVO. Scalebar, 300 μm.

Table 5. Central retinal vein occlusion (CRVO) from AOSLO-FA.

Using the Tree-Network Composite option of VESGEN, overall vascular patterning of capillaries, arteries, and veins in a normal retina of a healthy subject and in the two retinas of a patient with unilateral CRVO were compared by two parameters, the fractal dimension (D_f) and total vessel length density (L_v). Results for vessel branching generations of one to eight (G_v1–8) were further separated into large (G_v1–3) and small vessels (G_v≥4) by the software grouping function. To examine the uniformity of the capillary network, the AVS were calculated as mean area ± SD.

Parameter	Symbol	Normal retina	Fellow retina to CRVO	CRVO retina
Fractal dimension	D f	1.629	1.568	1.521
Length density
All vessels	L v	174.2	111.5	89.2
Large vessels	L v1–3	9.7	11.5	10.0
Small vessels	L v≥4	164.5	100.0	79.3
Number of AVS		1,855	821	526
Area of AVS	x±SD	1,352±4,712	2,555±8311	4,495±26,132
Ratio of AVS/ROI		0.664	0.708	0.785

Dimensions for L_v, ×10^–3 μm/μm² and area of AVS, μm² (D_f, dimensionless).

As a consequence of occlusion of the central retinal vein, overall vascular density by D_f and L_v, together with the uniformity of the AVS, decreased considerably in the CRVO retina compared to fellow and control retinas. Results of this preliminary study indicate by all confirming measures a substantial loss of vascular space-filling capacity in the CRVO retina. As quantified by generational analysis of L_v, the major vessel loss appears to be of the more vulnerable small vessels. The area of the CRVO avascular fovea appears considerably enlarged, indicating further loss of capillaries. By D_f, L_v and L_v≥4, the retina of the fellow eye in this middle-aged subject displayed an intermediate status of vascular/capillary capacity compared to the normal, healthy retina of the young adult subject, as is consistent with another AOSLO-FA study on fellow eyes [58].

Vascularized Tissue-Specific 3D Bioprinting from 2D Vascular Pattern

The assignment of branch specificity to the material and cellular composition of 3D bioprinted vascular constructs is described as a new application of VESGEN. Generic vascular patterns have been printed before [59,60], but without consideration to the longitudinal variations in branch composition. To illustrate the 3D printing of a vascular branching structure, a sector of an isolectin-labeled rat retina analyzed by VESGEN (Fig. 8a) was used to establish the branching generations (Fig. 8b–c) that was reduced to three vessel groups (large, medium and small, Fig. 8d). The image was converted in a printable .stl format (Fig. 8e), and direct-write printed [59] with a surrogate bioink (Fig. 8f).

Fig. 8. Branch-specific ‘direct write’ 3D printing of a vascular region from a rat retina.

(a) Original fluorescent image. (b) Same image after inversion, segmentation, and smoothing (scalebar, 100 μm). (c) Determination of branching generations by VESGEN (legend). By the analysis, not all branching generations were present, such as G₂ and G₃. (d) Same image after grouping of generation classes into large (G_1–3, red), medium (G₄, green) and small (G_≥5, blue). (e) CAD conversion of the same field into a .stl printable file. (f) Direct-write printing of the vascular pattern with a bioink surrogate.

For this proof-of-concept application, the use of VESGEN is demonstrated by the workflow leading from a 2D vascular image to a printable vascular network with region specificity in vascular dimensions and cellular composition. This method is currently being used to print vascular constructs of the proper size in both the pre-capillary and capillary range, with domain-specific proportions of vascular smooth muscle and endothelial cells, embedded in bioinks with a fine-tuned composition that is adequate for the respective vascular branch (work in progress). To this end, the bioprinter’s printhead will be loaded with bioinks containing different cell types and different ingredients. For printing arterioles, for example, a higher proportion of smooth muscle cells to endothelial cells will be added than for smaller vessels that may either lack smooth muscle cells altogether or be replaced with pericytes.

Moreover, in addition to vascular synchronization, VESGEN could further support the evolution of bioprinting from the current stage in which the bioinks are mainly generic biocompatible hydrogels such as alginate, hyaluronate, and/or gelatin [61] or fibrin [62], to the next level where the bioinks will be prepared from extracellular matrices of specific tissues or vascular beds [35]. In this context, the ability to co-register the original microvascular image with the bioprinted pattern is essential, in order to apply a phenotypically matched bioink appropriate for the particular (micro)vascular branch that potentially would also contain segment-specific immobilized growth factors. Such advances will be a significant technological advantage for creation of anatomically correct and physiologically relevant vascular constructs for regenerative tissue engineering.

Discussion

This overview of diverse in vivo and medical models analyzed by the VESGEN analysis demonstrates that mapping and quantifying vascular patterns can provide insightful readouts and integrative biomarkers of complex, multi-scale signaling for physiological, pathological, and therapeutic research discovery. Vascular patterning may offer tissue-specific templates for the 3D bioprinting of regenerative tissues, organs, and replacements, such as occurs in normal development. Other models analyzed elsewhere by VESGEN include infant retinopathy of prematurity in mice [8], developing and pathological coronary vessel models [63,8] and VEGF-controlled lymphangiogenesis [13]. Globally available since 2019, the software is now being used independently in other laboratories outside of NASA that includes previous tumor angiogenesis studies [26] and now human retinopathy and 3D bioprinting.

The analysis by VESGEN is based on vertebrate vascular branching rules established by our research and others [21,64–68]. Vertebrate vascular systems are determined by complex interactions of numerous molecular signaling cascades with hemodynamic factors and specialized tissue architectures. The vascular system is the first to develop in vertebrates, and patterns of the extracellular matrix surrounding and defining tissue-specific vascular branching may be the primary determinant of tissue-specific architecture [69]. The pro- and anti-angiogenic effects of cytokine families such as VEGF, FGF, Insulin-like Growth Factor (IGF) and TGF-β are well established. Over 80,000 citations on VEGF alone now appear in PubMed [70]. In recent years, inflammation has received considerable attention as a unifying perspective on vascular remodeling. Initial studies with VESGEN in the avian CAM revealed that VEGF-165, bFGF, TGF-β1 and other regulators elicit unique ‘fingerprint’ or ‘signature’ vascular patterns [9,11,10,14, 13,63,15,8,32]. For example, bFGF stimulated only a robust dose-dependent angiogenic response. However, VEGF induced a bimodal dose-dependent response that switched from a physiological phenotype of normal angiogenesis to a pathological phenotype of excessive vessel dilation and leakage accompanied by activation of eNOS. Vascular branching in Danio rerio (zebrafish) would also be an interesting model for the scale-invariant VESGEN vascular analysis.

The role of physics in cardiovascular fluid mechanics has also been studied as a causative force in microvascular patterning and remodeling [23,71,45]. Substantial research has shown previously the contributions of blood pressure, pulsatile flow, shear stress, and cyclic shear stress to the vascular signaling of molecular response within normal, inflammatory, or other pathological contexts [72–76]. Results reported here on a vessel tortuosity phenotype of decreasing sinusoidal amplitude within a subset of DR patients, and induction by epithelial activator KGF of significant vessel tortuosity, further suggest that signaling by pulsatile blood flow is involved in the remodeling of vascular morphological pattern. We are often asked if VESGEN can analyze the fractal branching patterns of arterial river systems and neurons (Fig. 9) [77,78]. Whereas river and vertebrate vascular branching are determined by aqueous flow, the branching of neurons appears to derive from conduction of electrical charge and, therefore, more resembles the branching patterns of lightning. Rules based on electrical signaling would therefore be required to develop successful algorithms for neuronal fractals. We can only comment that the interactive roles of cardiovascular hemodynamics and fractal vascular branching have yet to be defined for the determination of microvascular morphology and remodeling.

Fig. 9.

Fractal patterning of self-similar branching in the aqueous transport of river systems and conduction of electrical charge in lightning Photographs of (left) arterial river branching in the Sundarbans coast at the Mouths of the Ganges by the Bay of Bengal [82] and (right) lightning at night [83] illustrate the contrasting physics of self-similar fractal branching in the transport of aqueous fluids such as rivers and biological vascular systems, compared to the conduction of electricity in lightning and neurons of nervous systems. Self-similarity is a fractal pattern whereby the characteristic pattern (in these examples, bifurcational branching) is repeated at increasingly smaller length scales. Unlike many mathematical fractals, fractals in biology and nature are irregular.

Other future advances would include addition of bioinformatic dimensions to VESGEN [16,79] by which patterns of molecular expression such as signaling receptors can be co-localized with vascular pattern. Non-vertebrate applications of VESGEN include vascular networks in genetically modified wings of Drosophila melanogaster (fruit fly) [17] and the fractal scaling of developing leaves in the angiosperm Arabidopsis thaliana (thale cress) [16]. These studies required some overriding of a basic branching algorithm of VESGEN. The rules for bifurcational vascular patterning in insects and angiosperms, designed for optimal transport of vascular fluids as in vertebrates, are both similar and different from that of vertebrates. As for neuronal branching, a fully automated analysis would require additional branching algorithms for VESGEN. Interestingly, the species-specific leaf venation patterning of angiosperms such as thale cress, maple, and oak is an accepted taxonomic species identifier. Leaf venation in angiosperms has sometimes been viewed as a major causative factor in the explosion in biodiversity during the Cretaceous period, because of increased efficiency in leaf photosynthesis and other vascular hydraulics compared to gymnosperms [80,81].

In general, relatively small sample sizes have been required to achieve statistical significance for studies with VESGEN, perhaps because of the large ‘sampling’ numbers of vessels in most clinical or experimental images (such as hundreds of vessels in the human retina). With the exception of the CAM model, experimental visualization of in vivo microvasculature is difficult, requiring specialized techniques such as vascular perfusion, sophisticated immunohistochemistry or genetic markers such as in Tie2-Cre transgenic mice. For medical studies, the human retina is a relatively accessible site for imaging by fluorescein angiography or other recent sophisticated methods such as ocular coherence tomography-angiography (OCT-A). Nonetheless, analysis by VESGEN currently requires the challenging manual segmentation (extraction) of a binary (black/white) vascular pattern from microscopic or clinical images using software such as Photoshop^® (Adobe) or ImageJ to provide the input image. Consequently, NASA is using AI/machine learning to develop the automated vascular segmentation (VESGEN v1.11) described earlier. Such segmentation is currently the subject of intense worldwide research for numerous vascular biomedical applications.

In conclusion, the remodeling of both normal and pathological vasculature offers an integrative, informative read-out of complex molecular and other physiological signaling. Re-normalization of the vasculature can be determined in part from the response of vascular pattern to therapeutic testing. As a research discovery tool, the recently released VESGEN software has been used productively to investigate vascular remodeling in human, rodent, avian, and other organisms, and now the retinas of astronauts, to help develop countermeasures against the Spaceflight Associated Neuro-Ocular Disorder (SANS) risk for long-duration space exploration and colonization.

Availability of data and material

Availability of data and material is subject to permission by the contributing principal investigators of the various studies.