Current understanding of subclinical diabetic retinopathy informed by histology and high‐resolution in vivo imaging

The escalating incidence of diabetes mellitus has amplified the global impact of diabetic retinopathy. There are known structural and functional changes in the diabetic retina that precede the fundus photography abnormalities which currently are used to diagnose clinical diabetic retinopathy. Understanding these subclinical alterations is important for effective disease management. Histology and high‐resolution clinical imaging reveal that the entire neurovascular unit, comprised of retinal vasculature, neurons and glial cells, is affected in subclinical disease. Early functional manifestations are seen in the form of blood flow and electroretinography disturbances. Structurally, there are alterations in the cellular components of vasculature, glia and the neuronal network. On clinical imaging, changes to vessel density and thickness of neuronal layers are observed. How these subclinical disturbances interact and ultimately manifest as clinical disease remains elusive. However, this knowledge reveals potential early therapeutic targets and the need for imaging modalities that can detect subclinical changes in a clinical setting.


| INTRODUCTION
The incidence of diabetes mellitus (DM) is increasing world-wide. 1 Diabetic retinopathy (DR) is a common complication for patients with DM and is one of the leading causes of blindness in the world. 1 Currently, the diagnosis and grading of DR is reliant on the visualisation of macroscopic structural changes on fundus photography, as defined by the Early Treatment Diabetic Retinopathy Study (ETDRS) reports. 2 The earliest of these changes seen on fundus photography is the microaneurysm, which is diagnostic of clinical DR. 2 Although imaging with fundus photography remains a cornerstone of DR management, its limitations are becoming increasingly apparent.4][5][6] By definition, these changes are subclinical and characterise what we term as 'subclinical DR' in this paper.Understanding these changes, which hold crucial pathogenic significance, will help elucidate early therapeutic targets and improve disease management.
The retina is comprised of three main cell types: vascular, neuronal and glial cells. 7These cellular populations form a cohesive functional entity known as the neurovascular unit (NVU). 7Given the absence of autonomic innervation to the human retina, one major function of the NVU is to facilitate autoregulation of retinal blood flow in response to rapidly fluctuating metabolic demands of the neurons. 8A pathologic alteration to any one of these cellular components is therefore likely to affect the function of the NVU, and the retina as a whole.By marrying findings from histology and high-resolution clinical imaging, we can understand how cellular components are altered and how this may affect retinal function in subclinical disease.
Experimental animal models and clinical electroretinography (ERG) have shown that retinal vasculature and neurons exhibit altered function in subclinical disease. 3,4hese present as measurable differences in blood flow rates, ERG amplitudes and latencies in eyes that macroscopically appear 'disease-free'. 3,4Histological imaging of animal and human retinae similarly reveals subclinical structural changes affecting vascular, neuronal and glial cells. 5,6There is a need to elucidate these key subclinical changes, both functionally and structurally, in each of the cellular populations of the retina.In this review we outline the current understanding of subclinical DR guided by histology and high-resolution clinical imaging.

| FUNCTIONAL CHANGES IN SUBCLINICAL DIABETIC RETINOPATHY
The existence of subclinical changes to retinal neurovascular function are supported by in vivo experimental animal models and human clinical investigations. 3,4,9DR was classically thought to only affect the retinal vasculature, but it is now understood that neuroglial changes may occur concurrently or even precede vasculopathy. 10This is logically consistent with the understanding that vessels and neurons do not operate as independent structures, but are intrinsically connected through glial cells. 5The connection of these cells form the NVU, one function of which is the autoregulation of microvascular blood flow. 8Autoregulation maintains retinal homeostasis by titrating blood flow to fluctuating metabolic demands in response to local factors such as carbon dioxide and nitric oxide. 8,114][5][6] Therefore, it is unsurprising that measurable disturbances to retinal function are evident even in the earliest stages of disease.

| Changes to vascular function
In vivo animal experiments were some of the first to demonstrate change to retinal haemodynamics in subclinical DR. 3,[12][13][14] The approach of these studies was to measure retinal blood flow in very early diabetic animals using specialised techniques.Cringle, Yu and colleagues used hydrogen clearance polarography to report a significant increase in both mean retinal blood flow and blood flow variability in 5-week streptozotocin (STZ) induced diabetic rats. 3Meaning these early diabetic rodents had a wider range of retinal blood flow rates, which, on average, were higher than those of control rats.Similar results were detected with other techniques using radioactively labelled albumin and microspheres in comparable early STZ models. 12,13In vivo imaging with fluorescein angiography (FA) has also been used as a surrogate measure of blood flow in 1-week STZ-induced diabetic rats and 1-month Goto-Kakizaki rats, which mimic type 1 and 2 DM respectively. 15,16These studies found increased mean fluorescein circulation times, implying a reduction to retinal blood flow in subclinical disease. 15,16heir findings oppose the reports from Cringle et al. 3 and Tilton et al., 13,17 which found a mean increase in retinal blood flow.Comparatively, these reports used a more direct technique of blood flow measurement than FA. 3,13,17The inconsistency may be due to the different techniques employed or varying levels of hyperglycaemia that the rodents were exposed to across investigations.Fluorescein injections for example are administered through the venous circulation and transit times are therefore influenced by cardiovascular and haemodynamic factors separate from the retinal circulation such as carotid stenosis and cardiac output. 18Hydrogen clearance polarography by contrast is administered through intra-arterial cannulation of the carotid artery, bypassing the venous circulation. 3Furthermore, one study found only a weak correlation between FA mean transit times and microsphere blood flow measurements. 19Nevertheless, it is clear that retinal haemodynamic changes occur in subclinical disease.Interestingly, these vascular disturbances appear to change nonlinearly throughout disease progression.Burns' research group utilised adaptive optics scanning laser ophthalmoscopy (AOSLO) in control and diabetic patients with and without retinopathy, finding an initial increase and then progressive decline in retinal blood flow with severity of clinical disease. 20,21his dynamic alteration may reflect initial adaptations in the microvasculature that eventually decompensate with more advanced disease.][23][24] Studies have consistently shown that there is a measurable alteration in retinal blood flow at the earliest stages of subclinical disease.The histopathologic changes to key vascular structures that may result in these haemodynamic disturbances are discussed in Section 3.1.

| Drug vasoreactivity
The vasculature of very early diabetic animal models show remarkably different drug reactivity to that of controls.Alder and colleagues' ex vivo technique of isolated rat arteriole perfusion discovered that within 4 weeks of STZ-induction, the vascular response to vasoactive compounds was significantly altered. 25Specifically, they found an increased reactivity to catecholamines and reduced reactivity to serotoninergic agents. 25In vivo rat studies following the same principle, but using less precise fundus photography derived measures of vessel diameter, found similarly altered vessel reactivity in early diabetes. 26These results imply that alterations to vascular control mechanisms, especially with regard to the endothelial cells and contractile proteins, are perturbed even at this very early stage.

| Changes to neuronal function
Over the two last decades there has been growing acceptance that neuronal dysfunction is a very early feature in DR pathogenesis. 48][29][30] This variety of clinical tests captures the multitude of different neuronal populations that may be involved in subclinical disease.
Visual field testing reveals that patients with subclinical DR are more likely to have field defects than non-diabetic controls. 27,28Furthermore, over time, there is progression of visual field defects even without clinical progression of disease, as per the ETDRS classification. 2,27,28Contrast sensitivity and colour discrimination are similarly diminished in subclinical disease. 29,30A limitation of psychophysical testing is that central nervous system (CNS) processing plays a significant role in the perception of testable stimuli. 31herefore these measures may not strictly capture retinal function only.Theoretically, in order to measure solely retinal function with visual stimuli testing, an intact visual cortex and extra-ocular neural pathway is required.However, it is known that the CNS and, by extension, the visual cortex, are not immune to the insults of DM. 32 ERG itself measures retinal neuronal activity through quantification of micro-voltage readings under differing light conditions. 4Recordings from ERG methods are assumed to be independent of the CNS, and in this way quantify the functional output of retinal neuronal populations accurately.It is well established that diffuse ERG abnormalities are present in eyes with severe DR and reports have shown certain changes are also present in subclinical disease. 4,33,34ERGs of subclinical patients show that many neuronal cell populations are affected functionally, as they are structurally in diabetic animal models. 4,35The standard surveys of dark-and light-adapted a-and b-wave ERG responses, which target rod and cone populations, show increased latencies of light-adapted responses and reduced amplitudes of dark-adapted responses. 4Inner retinal neurons, in particular, appear to be the first neuronal population to show functional change in subclinical DR. 34,36,37 Oscillatory potentials that reflect bipolar and amacrine cell interactions, show consistent and reproducible amplitude reductions in early subclinical DR. 34,36,37 Photopic negative response (PhNR) ERG, which assesses retinal ganglion cell (RGC) function, shows significant impairments in subclinical DR. 38,39 These results are consistent with anatomical knowledge that the inner retinal layers, including the ganglion cell layer (GCL), are thinned in early DR. 40More recently, modified contrast testing, which quantifies the function of specific ON RGC sub-types, has demonstrated a loss of sensitivity in subclinical disease. 41,42ased on the principal concept that neurons and vasculature do not operate independently, stimuli-response experiments have been designed to estimate NVU functional status in subclinical DR.Lecleire-Collect and colleagues investigated this aspect using flicker-induced vasodilation responses in tandem with ERG, finding both components to be impaired in subclinical DR. 43 Flickerinduced experiments are considered measures of endothelial and contractile cell responses to light stimuli which are detected by photoreceptors and communicated by neuroglial cells. 44Lasta and colleagues found slightly differing results in early DR patients. 45They saw flicker light response compromise without pattern ERG dysfunction, implying that neurovascular coupling may be affected even before neuronal dysfunction in early disease. 45hese findings and those in Section 2.1 suggest that every component of the NVU is functionally compromised in subclinical DR.Section 3 will describe the histopathological patterns of vascular, neuronal and glial alterations that may be the anatomic reflections of these functional changes.

| HISTOPATHOLOGIC FINDINGS OF SUBCLINICAL DIABETIC RETINOPATHY
Histological examination of animal and human retinae provides insight into the earliest pathologic alterations that may be undetectable by current high-resolution clinical imaging.Understanding these changes reveal possible mechanisms underlying the functional alterations evident in subclinical DR.Histopathologic changes are evident in all three major retinal cell types: the vascular apparatus, neuronal cells and glia.

| The vascular apparatus
The vascular apparatus of the retina comprises the major arteries and veins, arterioles and venules, capillary beds and constituent cellular components that create a dynamically functioning circulatory system. 5Cellular components include endothelial cells, pericytes and vascular smooth muscle cells (VSMC). 5Changes to the general architecture of retinal vasculature and variations in these components in subclinical DR will be discussed.

| Changes to vascular architecture
Clinical DR is characterised by measurable architectural changes to the vasculature in the retina.For example, capillary density loss is progressive and occurs both peripherally and centrally in DR. 46 Histological studies in subclinical DR reveal more subtle architectural changes.It is understood that of the three vascular plexuses in the human macula, there is a typical three-dimensional arrangement of connectivity between each layer. 47An and colleagues, through perfusion labelling of human donor retina, found that diabetic patients without DR have an increased number of direct arteriole connections to the intermediate capillary plexus (ICP) compared to controls. 47,48Given that within the macula, the deep capillary plexus (DCP) is almost exclusively supplied by the ICP, this change presumably facilitates greater redistribution and shunting of blood towards the deeper vasculature. 47,49,50This may reflect a disproportionate increase in energy demand of the inner nuclear layer (INL) and inner plexiform layer (IPL) in subclinical disease. 47,49,50his altered arrangement may not influence total serial resistance of the retina, but due to an increased number of vessels in parallel, will affect individual vessel blood flow rates.This effect may contribute to changes in small vessel blood flow rates seen using AOSLO. 20he same paper by An et al. 48discovered reductions in capillary density on histological specimens were only present in those with clinical disease, implying ischaemic mechanisms may not be at a level significant enough to cause capillary bed loss in subclinical DR.However, this is in contention with results of clinical angiographic studies that do find subclinical capillary density loss. 51This disagreement is discussed further in Section 4, but may be apparent because of key distinctions between histologic perfusion labelling and clinical imaging techniques.Perfusion labelling, which fixes and stains all patent or semi-patent retinal vessels, does not emulate imaging under physiologic perfusion, which is influenced by constant autoregulation.In more advanced disease, capillary non-perfusion is grossly evident, the mechanisms of which are not entirely understood but may include endothelial cell hypertrophy and proliferation, leukocyte adherence, thrombosis and shunt vessel formation. 5,52here is evidence that some of these processes may begin to occur in the subclinical stages of disease.

| Endothelial cells
Endothelial cells create the internal lining of retinal vessels and form the tubular structure of capillaries. 5etween adjacent endothelial cells are tight junctional proteins that contribute to the formation of the internal blood retinal barrier (BRB) in conjunction with pericytes, VSMCs and glia. 5Endothelial cells, being in direct contact with pathogenic factors in the circulating blood, are naturally implicated in early DR pathogenesis.For instance, they express receptors for key inflammatory cytokines such as vascular endothelial growth factor (VEGF), tumour necrosis factor alpha (TNFα), interleukin 6 (IL-6) and targets for adhesion molecules. 5,53Dysfunction of endothelial cells is one of the earliest changes in clinical DR and is thought to contribute to many pathologic features such as leakage, vessel occlusion, leukostasis, thrombosis and microaneurysms. 5,54Endothelial cell dysfunction in subclinical DR is strongly supported by experimental vasoreactivity and flicker light studies showing altered vessel responses in early disease. 25,43,53n the clinical setting, this dysfunction can be indirectly quantified through tests assessing reperfusion behaviours of vessels, such as venous occlusion plethysmography, or by measuring serum levels of surrogate markers of dysfunction, such as C-reactive protein (CRP) and cluster of differentiation 40 (CD40). 55,56In contrast, histological evaluation of endothelial cells is one of the few methods that allows direct visualisation of cellular structure and pathology over indirect measures of function.
Historically, one histological finding in the early diabetic retina is the occurrence of acellular capillaries. 57,58hese are thought to be capillaries which have suffered endothelial cell loss and remain as a basement membrane shell. 57,580][61][62] Analysis of human donors show that almost all microaneurysms express an intact endothelium. 59,60For this to occur, proliferation of endothelial cells prior to or concurrent with microaneurysm formation, is essential.Therefore, endothelial cells must have the ability, or even the propensity, for proliferation in subclinical disease.Despite this, one study using propidium iodide, a marker specific for both apoptotic and necrotic cells, found evidence of endothelial cell loss as early as 1-week post STZ-induction of diabetes. 58The pathogenesis of this endothelial cell death has been linked to microglial inflammation and the expression of intercellular adhesion molecule 1 (ICAM-1), which has also been suggested to be important in the lifecycle of microaneurysms. 58,60,63istologically, this is seen as an increase in adherent intraluminal inflammatory cells in diabetic retinal vasculature associated with endothelial cell loss. 58Earlier studies using STZ-induced diabetic rats found evidence of endothelial cell proliferation after 6 weeks and a decreased number of endothelial cell nuclei thereafter. 61,62This finding shows both endothelial proliferation and cell loss may occur in the same animal model.From our histological analysis of advanced DR in human donor eyes, we see that proliferation and/or hypertrophy of endothelial cells, in non-aneurysmal locations, contributes to vessel occlusion. 5This finding is most obvious in advanced disease but may be occurring subtly in the subclinical retina, at levels not yet severe enough to cause focal ischaemia. 5o rationalise these paradoxical findings of proliferation and cell loss, it is possible that acellular capillary occurrence may be the endpoint of focal endothelial cell hypertrophy and/or hyperplasia that eventuates in lumen narrowing and fine vessel occlusion.The occlusion translates to vessel non-perfusion, endothelial cell death and finally acellular capillaries as reported in earlier histologic studies. 5,57,58These cellular mechanisms, contributing to vessel occlusion in DR and possibly present in subclinical DR, represent an important yet poorly understood aspect of disease.
Endothelial cells of retinal vessels contain a cytoskeleton which is seen histologically as intracellular fibres. 5hese fibres stabilise cell structure and junctional proteins, thereby supporting integrity of the vessel and internal BRB. 5,64Accordingly, these fibres are greatest in density at high-pressure locations such as arteries and arterioles. 64Changes to the distribution and density of these intracellular fibres has been documented in histological studies of subclinical DR. 64 In early STZ-induced diabetic rats, the endothelial cells of capillaries expressed filamentous actin (F-actin) more diffusely compared to controls, with notable areas of stress fibre drop out that were closely associated with sites of microvascular leakage. 64Our laboratories unpublished data in human control and diabetic donors demonstrate similar changes to F-actin distribution and loss associated with focal leak in clinical DR (Figure 1).These findings show that even with a preserved endothelial cell count, a breakdown of the internal BRB may occur.Microvascular leakage likely occurs because F-actin filaments support the function of cell to cell junctional complexes which contribute to the internal BRB. 65These alterations are mirrored by results from cardiovascular studies which demonstrate abnormal morphological alignments of cytoskeletal fibres in response to mechanical stressors in aortic cell cultures. 66he importance of this parallel is that changing mechanical or haemodynamic states, as is known to occur in subclinical DR, are potentially the cause or consequence of stress fibre changes. 67

| Pericytes
Pericytes are cells found on the abluminal surface of lower order retinal blood vessels which function to regulate blood flow, reinforce the internal BRB and modulate signalling pathways. 5Similar to endothelial cells, loss of pericytes in the form of 'ghost cells', is one of the main features seen in early histological studies and is considered a 'hallmark' of DR. [68][69][70][71] Loss of pericytes is not unique to the retina and is documented to occur in the kidneys, brain and heart of diabetic patients. 724][75][76] Pericyte ghosts were first discovered almost six decades ago when Cogan and Kuwabara used trypsin digests to isolate retinal vasculature. 77It is possible that pericyte ghosts, which imply pericyte death, are instead artefacts of this trypsin digest technique.In light of this, theories of pericyte migration over pure pericyte death have been suggested as the primary contributing factor to the changes observed in pericyte populations. 78,79Support for this is from histological observations in animal models by Pfister and colleagues of reduced pericyte count on straight capillaries, but an increase in pericyte count elsewhere. 79hese changes, however, were only studied after 6 months of experimental diabetes, so implications for very early disease are not certain.In our laboratories experience of perfusion labelling hundreds of human retinae, pericyte counts appear preserved or only mildly diminished in subclinical disease, though confirmation is required with quantitative analysis.The process of pericyte migration and therefore early disease pathogenesis is influenced by angiopoietin-2 (Ang-2), a vasogenic molecule increased in DR and recently implicated as a therapeutic target for diabetic macular oedema (DMO). 80Interestingly, the expression of Ang-2 and its receptor have yet to be precisely characterised in the human retina, knowledge of which may have implications for understanding pericyte migration and treatment of DMO.
Further histological evidence for pericyte dysfunction is found in the close relationship pericyte absence has as a precursor for microaneurysm formation. 46,60,70In other words, pericyte compromise or migration may occur in some cases prior to the emergence of clinical disease.An et al. 60 found, from histological analysis of human donors, that only 39% of the 636 microaneurysms due to diabetes harboured pericytes.Electron microscopy investigations show similar findings of selective pericyte losses in short duration diabetic human donors and animals, prior to endothelial and aneurysmal changes. 81istologically, pericytes are challenging to accurately label and visualise and the reasons for this are twofold.First, there is no single marker that exclusively stains pericytes; second, the expression of molecular targets are heterogenous across pericyte populations and vascular layers.For instance, neuron-glia antigen 2 (NG2) is a marker commonly used to label pericytes, but is also expressed in neuron and glia cells. 72In our experience, NG2 stains relatively non-specifically, limiting its utility as a reliable pericyte marker.Some pericytes express alpha-smooth muscle actin (αSMA), the same contractile protein present in VSMCs, the distribution of which is largely localised to arteriole inflow sites and changes with diabetes (Figure 2). 5 The implications of this is that multiple histological markers are required to accurately visualise pericytes across all vascular layers.

| Contractile proteins
Contractile proteins in the retina are numerous and can be found within VSMCs, pericytes and endothelial cells of retinal vessels. 5These structures perform many important functions including maintaining integrity of the internal BRB, signalling for normal vascular formation and regulation of blood flow to meet metabolic demands. 82In the previous sections we have provided evidence that haemodynamic flow and neuronal function is altered in subclinical disease, it is therefore likely that expression of these vasoregulatory proteins are associated with functional changes in blood flow.
VSMCs express the largest concentration of contractile protein in the retina. 81,83Similar to endothelial cells and pericytes, loss of VSMCs is thought to occur in early disease. 81,83This is consistent with knowledge that VSMCs and pericytes have similar embryological origins and presumably share susceptibilities to hyperglycaemia. 81,84I G U R E 1 Endothelial cell filamentous actin (F-actin) stress fibre distribution in arterioles of age-matched control, subclinical diabetic retinopathy (DR) and DR human donor retina.All images are taken at the distal arterial arcade.Non-diabetic donors (A) demonstrate an organised distribution of F-actin fibres (green arrows) throughout the endothelial cell which runs perpendicular to flow in the vessel.In subclinical DR (B) there is an increased expression of F-actin stress fibres expressed diffusely and more densely (green arrows) within the endothelial cell compared to control, with no evidence of leak in this specimen.In clinical disease (C) there is loss of F-actin fibres intracellularly (teal arrows) with evidence of focal leak associated with this stress fibre loss (yellow arrows).Vascular smooth muscle cells are also visible (white arrows), these smooth muscle actin fibres are external to endothelial cells and run perpendicular to blood flow.Scale bars = 30 μm.
Within VSMCs, select pericytes and some endothelial cells is the contractile protein αSMA, which is the major contractile protein found within the retina. 56][87] The expression of αSMA has been described with histological examination of F I G U R E 3 Alpha smooth muscle actin (αSMA) expression in parafovea of age-matched non-diabetic control retina (A) and diabetic retina without retinopathy (B).αSMA is predominately expressed by vascular smooth muscle cells of arterioles (a) and venules (v) in the superficial vascular plexus (SVP).Extension of αSMA from the arterial aspect into lower order arterioles and capillaries of the SVP is demonstrated by magnified insets (C) and (D) and is mainly expressed by pericytes not visible at this magnification (Figure 2).In the subclinical DR specimen, there is greater extension of αSMA into lower order arterioles and capillaries further in the SVP (Inset D).In subclinical DR and DR, αSMA is expressed in progressively greater amounts deeper into the intermediate and deep capillary plexus not visible here (Figure 4).There is also greater staining intensity of αSMA on the subclinical DR venule in this specimen compared to control.Yellow -αSMA.Scale bars = 50 μm.
human retina in the context of DR and subclinical DR. 48,88,89 Briefly, in non-diabetic controls, αSMA is expressed most greatly in the arterial and capillary circulations by VSMCs and endothelia of the superficial and intermediate vascular plexus (Figures 2 and 3).In the DCP of non-diabetic retina, there is minimal αSMA expression by pericytes and endothelium.In subclinical DR, there is increased αSMA expression in both pericytes and endothelial cells, which extends into the DCP and progresses with disease severity (Figures 3  and 4). 48Implications of VSMC loss and redistribution of αSMA, along with greater vascular connections into the ICP and DCP, may suggest that a greater autoregulatory capacity or functional hyperaemic effect is conferred to the deeper vasculature in subclinical DR.This observation, along with the changes noted in endothelial cell F-actin, may be cause or consequence of the changes in blood flow and neuronal function seen in studies of subclinical DR. 3,5,20,90 In DR, this mechanism may decompensate and progress to a pathologic level, or in the context of grossly altered haemodynamics such as ischaemia, result in the progressive retinal blood flow decline seen on clinical testing. 20Furthermore, many features of DR have a predisposition towards the INL and outer plexiform layer (OPL), which is the territory supplied by the DCP.Clinical examples of this spatial correlation include ERG dysfunctions localised to bipolar cells and paracentral acute middle maculopathy (PAMM). 5,36

| Summary of vascular histopathologic changes
The sum of these histopathological changes to the vascular apparatus in subclinical DR carries several implications.Clear disruption to the functioning of the internal BRB occurs, evidenced through loss and/or migration of pericytes, loss of VSMCs and endothelial cell dysfunction.Organised alterations to the architecture of vascular flow networks and contractile vasoregulatory components is evident.Both endothelial cell stress fibres and αSMA demonstrate a very organised redistribution towards the DCP in both human and animal models. 48,64This change may confer greater redistribution of flow and capacity for functional hyperaemia to the ICP and DCP.With these layers supplying highly metabolic neurons and synapses, the changes may be cause or consequence of neuroglial dysfunction.Without the appropriate autoregulatory capacity, this greater flow may predispose retinal neurons to cellular damage.As will be discussed in subsequent sections, it is clear that neurodegeneration is a prominent feature of subclinical disease.However, there is a paucity of evidence regarding the true timeline of these histopathological vascular observations in the context of neuroglial changes.

| Histopathologic changes to retinal neurons
From functional studies discussed in Section 2, it is clear that change to retinal neuron function is present in subclinical disease.][93] The neuroretina is organised in layers of differing neuronal populations, each of which may have particular susceptibilities in the context of diabetes.In clinical DR, it appears RGCs, the innermost neurons of the retina, are particularly vulnerable to injury. 62][93] In contrast, similar animal models documented no RGC loss after 6 weeks of diabetes, but rather a significant amount of astrocyte reactivity, axonal loss and optic nerve myelin changes. 94This pattern of optic nerve degeneration is consistent with rodent models reporting reductions in myelinated fibre size, the same effect that is documented on peripheral nerves in diabetic neuropathy. 95,96One further study found RGC loss is highly dependent on glycaemic control, with the same report finding no RGC loss in mice even after 10 months of diabetes. 97A comprehensive study by Sohn and colleagues investigated early neurodegeneration using optical coherence tomography (OCT) and histology in humans and animals. 98The histological investigation found significant nerve fibre layer (NFL) thinning but no RGC density difference in diabetic human donors with minimal or no retinopathy. 98In STZ-induced mice they found similar patterns to previous animal studies with RGC loss absent at 6 weeks but present at 20 weeks. 98nterestingly, they noted NFL and RGC layer thinning on OCT on the same animals as early as 6 weeks, implying layer thinning may precede actual cell loss.The authors noted these changes to precede decrements in pericyte density.Other neuronal populations have also shown alterations at this early stage of disease.Park and colleagues reported degenerative changes to photoreceptors at 1-week, mild RGC and photoreceptor loss at 4 weeks, amacrine and horizontal cell necrosis at 12 weeks and significant photoreceptor apoptosis after 24 weeks of STZ-induced diabetes. 35Gastinger et al. 99 also reported amacrine cell loss in the early stages of diabetes.
The culmination of these studies support neurodegeneration as a key feature of subclinical disease.Specifically, inner neuronal cell loss occurs in subclinical DR and there is evidence that axonal changes, reflected by NFL thinning, may develop even earlier. 35,981][102] In other words, in early diabetes, the neurodegenerative mechanisms causing transduction defects may not yet be severe enough to manifest as histologically-detectable structural change.This helps to explain the contrast between relatively inconsistent histological findings and the consistent functional changes seen in ERG studies.The compromise to visual pathways, like those documented in the optic nerve, may be further contributing to visual field detriments seen in subclinical and early DR. 27,28,94,95ery few studies have investigated these neuronal changes in conjunction with alterations to vascular components, such as, contractile proteins, pericytes or endothelial cells.These neurovascular changes both appear to develop within 1 to 6 weeks post diabetic induction in STZ rodent models. 35,64,81Given the implications that redistribution of contractile proteins and endothelial cell F-actin may have for blood flow to the inner retinal layers, future investigations should aim to understand expression of these vascular components in the context of neurodegeneration.Glia cells are the third major cell type of the neurovascular triad and functionally link vessels and neurons; as such, they are key to elucidating the relationship between these features of subclinical DR.

| Glia and gliosis
Like the CNS, the retina must receive a continuous supply of blood to match the metabolic demands of neurons and maintain homeostasis. 8Neuronal activity modulates local blood flow; a response termed functional hyperaemia. 44In the retina, glial cells are intercalated between vessels and neurons to directly communicate changes in neural activity to contractile units surrounding blood vessels. 103,104Impairment of glia and thereby the NVU, may precede the established clinical and morphometric microvascular changes of DR. 91,105,106 Hallmarks of DM induced retinal neuroglial degeneration includes reactive gliosis, diminished neuronal function and neuronal apoptosis. 91,105,106These changes have been observed before the presence of microaneurysms in human donors and animal models. 91,105,106At present it is not known whether neuronal apoptosis or reactive gliosis occurs first in this neurodegenerative process.Reactive gliosis, characterised by proliferation or hypertrophy of glial cells, as is seen in CNS injury, may damage retinal neurons and contribute to neurodegeneration and microvascular disease. 107

| Astrocytes
][110][111] Astrocytes are typically stellate in shape with long processes that ensheath blood vessels of the SVP and support nearby axons of RGCs. 111A second population of oval shaped astrocytes reside in the peripapillary capillary plexus of the NFL, with processes arranged parallel to axons and peripapillary capillaries. 111Astrocytes undergo reactive gliosis in response to polyetiological insults, such as trauma, ischaemia or inflammation. 112This response involves both structural and functional changes, however, the process may become maladaptive and constitute a pathogenic element.][114] Astrocytes have been shown to undergo gliosis in early diabetes (Figure 5). 115Vascular leakage associated with loss of astrocytic end-feet was seen in murine models as early as 3 weeks post diabetic induction. 116Astrocytic endfeet envelop blood vessels, forming part of the internal BRB, loss of which contributes to vascular leakage and impairment of NVU function. 116After 4 weeks of diabetes, gap junction protein and gene expression of connexin decrease, with subsequent astrocyte loss and morphological changes observed at 6 weeks. 117Connexins are integral to gap junction formation and cell-cell communication, the decreased expression in early diabetes reflects a reduction in gap junction function and diminished astrocyte communication. 117In a separate rodent study, astrocytes reduced their expression of GFAP after 8 weeks of diabetes, contrary to the upregulation seen in CNS astrocytes during injury. 118,119GFAP is an intracellular filamentous protein that serves many roles including structural support and cell motility. 118The change in expression of GFAP was postulated by the authors to negatively affect function of retinal astrocytes, and by extension, BRB integrity. 119In addition to morphological changes, astrocytes demonstrate an inflammatory role through the amplification and production of pro-inflammatory cytokines such as interleukin 1 beta (IL-1b), monocyte chemoattractant protein-1 (MCP-1) and VEGF. 120,121Together, these glial alterations may accompany or contribute to early vascular dysfunction in the form of internal BRB breakdown and retinal blood flow change. 22,46,122

| Muller cells
Muller cells (MC) account for approximately 90% of all retinal glial cells, spanning the entire retinal thickness, they form an anatomical connection between all orders of neurons and vasculature. 123The cell bodies of MCs are found in the INL, with their end-feet enveloping the neurons and blood vessels of the inner retina, forming the internal limiting membrane (ILM) at the vitreoretinal junction. 124Towards the outer retina they form tight junctions with photoreceptor cells which demarcate the external limiting membrane (ELM). 124Importantly, MCs also express various voltage-gated channels and neurotransmitter receptors, enabling the modulation of neuronal action potentials by mediating the extracellular concentration of neuroactive compounds. 125Because of this unique spatial arrangement and functional ability, MCs have many significant functions which can be divided into four categories: (1) uptake and recycling of neurotransmitters (glutamate, gamma-aminobutyric acid and hydrogen ions), retinoic acid compounds, and ions such as potassium K+, (2) control of metabolism and supply of nutrients to the retina, (3) regulation of blood flow and maintenance of the BRB and ( 4) aid light transmission to photoreceptors. 126,127Cs themselves are heterogenous; their structure, function and molecular expression vary depending on location within the retina. 128,129Centrally, MCs adapt to the unique microenvironment of the fovea by forming an inverted cone, providing optical and mechanical support. 130,131eripheral to the fovea, MCs are displaced and form highly elongated 'Z' shaped cells within the macula. 132MCs with these morphologies may be susceptible to injury in DR and contribute to the predilection oedema has for the macula. 132eripherally, MCs take on the shorter, more typical, columnar shape through all layers of the neuroretina. 132Besides structure, expression of important proteins also differ topographically throughout the retina.For instance, central MCs express more aquaporin-4 (AQP-4) whilst cluster of differentiation 44 (CD44) is exclusively expressed by peripheral MCs. 129,133It is likely these different MC subtypes are affected heterogeneously in subclinical DR, which may have implications for the distinct macular and peripheral disease that occurs in clinical DR.
Considering their strategic distribution, MCs may be one of the first glial cells to detect and respond to aberrations in the retinal microenvironment.A conditional MC ablation model demonstrated DR-like changes in MCabsent animals, such as photoreceptor apoptosis, BRB breakdown and neovascularisation. 134MCs themselves may further contribute to BRB disruption through the release of vasoactive molecules such as VEGF and TNFα. 135In the healthy retina, MCs either do not express GFAP or only express it minimally. 119,136However, under pathological conditions, MCs show upregulation of GFAP expression. 119,136In diabetic donors and animal models, MCs demonstrate this change at early stages of disease. 115,137,138Given the abundance of MCs, it is thought this upregulation of GFAP may be crucial for gliotic responses involved in glial scar formation, leukocyte infiltration, neurite growth and neovascularisation. 124,139ne study observed MC gliosis initially in the peripheral retina of 4-week STZ-induced diabetic rodents which then progressed centrally by 10 weeks. 117Other changes in early diabetes include compromises in glutamate turnover and uptake along with lesser expressions of glutamine synthetase, both of which are important for neurotransmitter regeneration. 138,140This coincides with reports of increased glutamate accumulation and neuronal excitotoxicity in the rodent diabetic retina. 137,141Glutamate accumulation that results from these early glial changes leads to abnormal synaptic transmission, affecting neuronal function. 137These MC changes likely contribute to ERG alterations seen in clinical studies of prediabetic and diabetic patients without DR. 4

| Microglia
Microglia are the resident immune cells of the retina and CNS. 128They are fundamental for normal retinal growth, neurogenesis, synaptic pruning, phagocytosis of cellular debris and balancing of inflammatory states. 142,143Like other major glial cell types, microglia are in close contact with neurons and synapses to stabilise and monitor neuronal function. 142,143Unlike macroglia however, microglia do not express GFAP. 128With knowledge that inflammatory states contribute to vascular occlusion in DR, microglial dysfunction likely has implications for both vasculopathy and neurodegeneration. 60,144ccordingly, microglia are known to be altered in clinical DR.When activated by metabolic stressors, microglia undergo hyperplasia and morphological change F I G U R E 5 Astrocytes in retinas of control (A) and 5-week streptozotocin rats (B).Reactive gliosis in early subclinical DR (B) is shown by increased expression of glial fibrillary acidic protein (GFAP) compared to control.Processes extending from astrocyte cell bodies are in close contact with blood vessels of the superficial vascular plexus represented in magenta and retinal neurons that are not visualised in this micrograph.Scale bars = 80 μm.
from a ramified to an amoeboid form, suggesting that activated microglia are highly proliferative and migratory. 145ccompanying this phenotypic alteration, activated microglia balance two opposing roles, triggering either proinflammatory or anti-inflammatory effects. 146,147In early DR, both responses occur concurrently to ameliorate inflammation and impede disease progression. 146,147However, as DR worsens, the homeostatic balance shifts towards a chronically pro-inflammatory response profile. 147This imbalance results from the release of proinflammatory cytokines, chemokines, caspases and glutamate, shifting the retina towards a neurotoxic microenvironment. 148In humans with moderate non-proliferative DR (NPDR), microglia proliferate and undergo morphological changes with clustering towards the inner retina, particularly around areas of vascular injury such as microaneurysms and intraretinal haemorrhages. 149Similar morphometric changes relating to an increased microglial density and activated phenotype have been observed in earlier models of 4-week STZinduced diabetic rats. 136,145Further rodent models observe structural changes with shortened dendrites and enlarged microglial soma prior to neuronal apoptosis, suggesting that microglial reactivity occurs early in DR pathogenesis. 150In the same study during the initial stages of diabetes, microglial reaction occurred approximately at the same time as ERG changes. 150Combined, these findings suggest microglial cytokine and morphological changes may occur concurrently with functional neuronal deficiencies but prior to changes in neurovascular structure.

| Optical coherence tomography
OCT revolutionised retinal imaging by enabling a noninvasive insight into the structure of the retina and choroid.In line with histological findings of inner retinal neurodegeneration, changes to the inner retinal layers have been seen in clinical imaging studies of diabetic patients without DR.A recent meta-analysis by Tang and Chan et al. 90 captured 36 studies with over 6000 subjects, investigating the use of OCT in this context.They found the NFL, GCL and IPL in the macula were significantly thinner in subclinical DR patients compared to control.In contrast, the NFL of the peripapillary area did not demonstrate any thinning in subclinical DR.This finding suggests inner neurons, or at least axons, in the macula may be more susceptible to neurodegeneration than other retinal areas. 90This meta-analysis considered only changes to the inner retinal neurons, however, there is evidence in animal models that photoreceptors are also implicated in subclinical DR. 35 Only a few clinical imaging studies have investigated photoreceptor death in early disease, the findings of which are conflicting. 151One 2016 study found significant photoreceptor layer thinning on OCT in subclinical disease, with the extent of thinning proportional to duration of DM. 152 In contrast, another report found thinning is apparent only in clinical DR. 153 In a recent study of prediabetic adult patients, no changes to GCL thickness were observed; however, the same patients interestingly demonstrated significant reductions in multifocal ERG amplitudes. 1001][102] Further improvements in OCT technology may provide the resolution needed to resolve some of this conflicting evidence and possibly detect earlier structural changes.

| Choroidal changes on OCT
With advancements in OCT technology such as enhanced depth imaging, the choroid has also been implicated in subclinical DR. 154 A meta-analysis by Endo and colleagues of 17 studies, capturing over 1000 diabetic eyes without retinopathy, concluded that the subfoveal choroid is significantly thinner in subclinical DR compared to controls. 154The mean thinning was 14.34 μm compared to controls, however a significant range of findings was present across the included investigations. 154Changes in choroidal thickness ranged between thinning of up to 71.4 μm and thickening up to 45 μm. 154,155The varied findings amongst these studies might be reconciled with the understanding that choroidal thickness is dependent on many variables, not all of which were controlled for in every study, such as age, sex, refraction, HbA1c, duration of diabetes and axial length. 154,156The mechanism of choroidal thinning remains unclear but may be related to findings in the diabetic choroid of non-perfusion and narrowing of vessels, similar to changes seen in DR. 154,157,158 These pathologic alterations to the choroid of diabetic patients could have a profound effect on the pigment epithelium and outer retina, contributing to photoreceptor losses seen in histological specimens. 35These changes should be interpreted in the context of the choroid's autonomic innervation, which implies a significantly different mechanism of flow regulation than that of the retina's exclusively autoregulatory nature.Practically, this means retinopathy like alterations in the choroid, as is seen in the limited number of histological studies available, may have different haemodynamic effects and physiological consequences than what similar changes may have in the retina. 158,159With this is mind, the current knowledge base of diabetic choroidopathy is very limited in comparison to the physiological importance of this tissue.Given the choroid is challenging to clinically image in high-resolution, future histological studies are required to facilitate understanding of choroidopathy in the context of DR and subclinical disease (Figure 6).

| Optical coherence tomography angiography
Optical coherence tomography angiography (OCTA) is a relatively new adaptation of OCT that enables rapid, non-invasive, depth resolved visualisation of retinal microvasculature. 159Previously, clinical imaging of capillaries required angiographic techniques using intravenous contrast dyes such as fluorescein or indocyanine green.The invasive nature of these tests made the study of microvasculature in vivo difficult, especially for subjects with no clinical evidence of DR.OCTA provides a significant improvement in the practicality of imaging retinal microvasculature, as such there has been a surge in studies using OCTA to investigate many retinal vasculopathies.3][164][165] Several papers only document significant capillary density reductions in the deep capillary plexus, however these results must be interpreted in the context of OCTAs limitations. 51,166,167][170] Regardless, these results are consistent with knowledge that DR changes have a propensity towards the deeper vasculature, perhaps due to the high oxygen demand of the OPL. 68,171,172One histological study of human donor retinae by our group found that capillary density reductions were only evident in clinical DR, whereas subclinical capillary density was not significantly different to controls. 48This conflict is reconciled with the understanding that visualising retinal vasculature with OCTA in vivo is very different to perfusion labelling of histologic specimens, where all patent microvasculature is fixed and stained.OCTA visualises perfused vasculature, above the threshold of device sensitivity, in physiologic and pathologic states with significant spatiotemporal variation. 8Meaning, the absence of capillaries on a single OCTA scan is not only the result of capillary non-perfusion, but may be an 'artefact' of normal or pathological perfusion variability.For this reason, volumetric registration and averaging of serial OCTA scans is required to obtain more accurate quantifications of vessel density. 173Despite this, registration and averaging are absent from many publications quantifying vessel density using OCTA.There is also the possibility that changes to vascular components discussed in Section 3.1, including architectural changes to connectivity between layers, result in more variable capillary blood flow rates. 47,48apillaries in subclinical disease may then be more likely to periodically drop below the devices threshold, thereby falsely emulating reductions in capillary density when measured by a single OCTA scan. 48ith these limitations in mind, several studies investigated both OCTA and OCT parameters in short duration DM patients.These papers find reductions in capillary density on OCTA but no accompanying changes to inner neuronal thickness on OCT. 166,174,175This suggests perfusion compromise, especially in the deep capillary plexus, may precede structural inner retinal neurodegeneration.By extension, a series of studies have investigated this paradigm with the inclusion of ERG.Ratra and colleagues, studying prediabetic and control patients, found no difference in FAZ area, capillary density or GCL thickness, but did record significantly reduced multifocal ERG amplitudes. 100Similar studies of diabetic patients without DR, detected variable capillary density reductions and multifocal ERG reductions, but no significant inner retinal thickness changes. 176,177aken together these studies of controls, prediabetic and diabetic patients without DR, that incorporate multimodal imaging and functional testing, suggest that neuronal dysfunction and perfusion changes may occur earlier than structural neurodegeneration.It is likely that profound glial dysfunction, discussed in Section 3.3, is contributing to the pathogenic picture of neurovascular changes detected by these studies.

| Functional vascular imaging
Functional vascular imaging refers to a modality that can assess retinal vasculature beyond structure and provide quantitative insights into function such as blood flow or F I G U R E 7 Quantification of macular capillary perfusion variability non-invasively with optical coherence tomography angiography (OCTA).Comparisons of consecutive OCTA scans (A and B) from the same eye taken seconds apart demonstrate variability in retinal capillary perfusion.Some capillaries are visualised in both scans (magenta arrow) and other capillaries are only seen on one scan (red arrow).Comparison of many consecutive OCTA scans allows the generation of a coefficient of variation (CoV) heat map (C), where the variation in perfusion within the macula is quantified and can be evaluated using an intensity scale.The variation in perfusion is estimated by the change in intensity of each pixel over the set of consecutive, intensity corrected and aligned OCTA scans.In panel C, vessels with low perfusion variability such as the feeding arterioles and venules are represented in blue whilst capillaries demonstrate greater perfusion variability.This technique may facilitate clinical assessment and quantification of subclinical changes occurring to vascular control mechanisms in subclinical diabetic retinopathy for improved diagnosis, monitoring and treatment response.
flow variability in vivo.Such a modality is important as it is evident that functional changes in the early diabetic retina occur prior to overt structural changes. 3,4,9Unlike neuronal changes, where ERG exists to clinically assess dysfunction, there is no established in vivo tests for vascular dysfunction beyond structural imaging.With compelling histological evidence that vasoregulatory components of blood vessels are significantly affected in early disease, such a technique may operate as a clinical marker for cellular injury to these vascular elements.Having a clinical standard for functional vascular imaging is important as it will establish a more complete pathogenic understanding of DR and facilitate improved screening and monitoring for diagnosis and treatment response.
OCTA can be argued to be a functional imaging modality as the angiogram is created from OCT reflectance decorrelations which are different between perfused and non-perfused areas. 159However, OCTA in this sense is limited by its capability to quantify true blood flow, in the context of current devices.In the last decade there has been development of functional imaging techniques in the literature using existing technologies.Burns and colleagues established an AOSLO technique leveraging the high spatial resolution of adaptive optics to visualise retinal perfusion in human subjects. 178They showed it is possible to detect capillary level perfusion changes in mild to moderate NPDR. 179Adoption of this technique is limited by equipment and relatively long scan time requirements which create barriers to entry for wider use.Laser Doppler velocimetry has been used by Grunwald et al. 180 and Nagaoka et al. 181 to measure blood flow rates in diabetes.However, these Doppler techniques operate on arterioles or venules which is in contention with knowledge that many subclinical changes occur at the capillary level.
Recently we have published a technique that quantifies perfusion variability using coefficient of variation (CoV) measurements derived from OCTA. 5,8,182,183Taking multiple consecutive OCTA scans of the macula, we can calculate the intensity variation of each pixel and quantify the spatiotemporal variation of retinal perfusion in vivo (Figure 7).This premise originates from histological evidence, discussed in Section 3, of compromise to key vasoregulatory components of the NVU that may pathologically affect variability in capillary perfusion.Further investigations have established the ability of this technique to discern areas of overt capillary ischaemia in cases of branch retinal vein occlusion. 183There are several advantages to this CoV technique, namely the ability to assess and quantify functional vascular changes in a non-invasive, rapid and practical manner, which can be immediately adopted by today's clinics.This technique, along with others like it, may serve as a biomarker for earlier diagnosis, progression or treatment response.

| CONCLUSION
DR continues to be a significant burden for patients and populations.It is no longer considered just a vasculopathy but rather a disease that affects every component of the NVU from its earliest stages.We have discussed the clinical, experimental and histological evidence that finds pathologic alterations early in the disease course, preceding clinical definitions.In short, changes to blood flow, vascular architecture, contractile proteins, pericytes, endothelial cells, glial activity and count, neuronal function, cell count and layer thickness, and choroidal thickness all occur in subclinical disease.However, there remains significant ambiguity regarding cause or consequence of these changes.For instance, there is an incomplete understanding as to how blood flow alterations and changes to structural and contractile proteins in blood vessels may be interlinked.There is emerging histological evidence that pericyte migration and endothelial cell proliferation occur in early disease.Furthermore, it is unclear as to which, if any, of these subclinical changes is the primary pathologic event and which may be occurring sequentially or in tandem.Despite this, it is likely that all of these subclinical alterations each contribute towards the ultimate manifestation of clinical DR.With this knowledge, it is not unreasonable to deduce that microaneurysm formation, which currently defines clinical disease, is, in fact, a relatively late declaration of disease.The importance of understanding subclinical disease is twofold.First, the findings provide direction for novel therapeutics targeting the earliest stages of disease, offering treatment before irreversible damage occurs.Second, the findings underscore the importance of establishing a new clinical imaging modality that captures the functional manifestations of these changes in vivo, thereby enhancing the management and outcomes of diabetic retinal disease.

ACKNOWLEDGEMENT
Open access publishing facilitated by The University of Western Australia, as part of the Wiley -The University of Western Australia agreement via the Council of Australian University Librarians.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.

F
I G U R E 2 Pericyte alpha smooth muscle actin (αSMA) expression in non-diabetic human perifoveal arteriole and capillaries.αSMA is expressed in retinal arterioles in the form of both vascular smooth muscle cells (A; white arrow) and pericytes (Inset B and C).Pericytes of the superficial vascular plexus (SVP) are more likely to express αSMA in low order arterioles (Inset B; yellow arrows) than capillaries (Inset C).Pericytes of the intermediate and deep capillary plexus (ICP; DCP) express minimal αSMA or do not express αSMA relative to pericytes of the SVP in the non-diabetic subject (Inset D, E and F).In the diabetic donor, extension of αSMA positive pericytes further and deeper into the capillaries of the SVP, ICP and DCP is evident in both subclinical disease and DR (Figures3 and 4).Three-dimensional reconstruction of this specimen demonstrating vascular layers is presented in Video S1.Scale bars = 5 μm.

F I G U R E 4
Schematic representation of αSMA distribution and changes in the development of diabetic retinopathy.Insets provide magnified views of regions of interest.(A) In the control group, αSMA (yellow dots) were predominantly localised within arteries, arterioles and capillaries on the arterial aspect of the circulation (red).For the venous aspect (blue), αSMA was localised to major venular junctions only.Within the connecting arterioles (a1) between the ICP and DCP, αSMA expression was found to terminate abruptly prior to reaching the DCP (inset I: black arrow).The locations of each vascular plexus are indicated in the retinal layers panel.(B) The DR-group showed additional αSMA expression along veins, venules and capillaries on the venous aspect of the circulation.Compared to the control group, there was more αSMA expression at venular junctions.In addition, αSMA expression along the a1 arteriole spanned its entire course and extended into the DCP (inset II; black arrow).(C) The DR+ group was characterised by microaneurysms and capillary dropout within the DCP (asterisks; inset III).Compared to the control and DR-groups, there was significantly more αSMA expression along veins, venules, and capillaries on the venous aspect.Within the DCP, αSMA expression was found distal to the a1 arterioles and at venular junctions (inset III; blue arrow).Reproduced with permission from An et al. 2022, https:// doi.org/10.1167/iovs.63.5.8.

F
I G U R E 6 Immunofluorescent staining of human choroidal vasculature in a human donor eye with diabetic retinopathy.Near complete staining of the choroidal vasculature (A) is achieved with perfusion delivery of antibodies via a single short posterior ciliary artery.Alpha smooth muscle actin (αSMA) is expressed on the arterial and venous aspect of larger vessels in Sattler's and Haller's layers.The dense choriocapillaris is visible across the specimen with arteries (a) originating from the same short posterior ciliary artery seen in magnified inset B. Histological methods such as this facilitate detection and understanding of pathologic changes across the entire vascular complex of the human choroid, which can be correlated with retinopathy status.Scale bars = 100 μm.