MRI of rod cell compartment-specific function in disease and treatment in vivo

Abstract

Rod cell oxidative stress is a major pathogenic factor in retinal disease, such as diabetic retinopathy (DR) and retinitis pigmentosa (RP). Personalized, non-destructive, and targeted treatment for these diseases remains elusive since current imaging methods cannot analytically measure treatment efficacy against rod cell compartment-specific oxidative stress in vivo. Over the last decade, novel MRI-based approaches that address this technology gap have been developed. This review summarizes progress in the development of MRI since 2006 that enables earlier evaluation of the impact of disease on rod cell compartment-specific function and the efficacy of anti-oxidant treatment than is currently possible with other methods. Most of the new assays of rod cell compartment-specific function are based on endogenous contrast mechanisms, and this is expected to facilitate their translation into patients with DR and RP, and other oxidative stress-based retinal diseases.

1. Introduction

The most prevalent photoreceptor in the mammalian retina is the rod cell. Rod cells play an essential role in both vision and health of other cells in the retina (Berkowitz et al., 2014; Bissig et al., 2013; Cingolani et al., 2006; Curcio et al., 1996; Dong et al., 2006; Du et al., 2013; Kassen et al., 2009; Kolesnikov et al., 2010; Komeima et al., 2006; Rogers et al., 2007; Shen et al., 2005; Usui et al., 2009b). Rod cells have two major and highly compartmentalized functions related to vision: light detection in its posterior outer segments and synaptic terminal neurotransmitter release at its anterior pole; the extracellular space surrounding the outer segments [i.e., the subretinal space (SRS)] is also a key compartment for maintaining a healthy dark current and visual cycle (Adijanto et al., 2009; Cao et al., 1996; Li et al., 1994b).

Diabetic retinopathy (DR) and retinitis pigmentosa (RP) are two major diseases of the retina characterized by irreversible anatomical changes, deterioration of vision, and ultimately blindness (Campochiaro et al., 2015; Kern et al., 2010; Punzo et al., 2012). Current treatment options are either destructive and sub-optimal for DR (e.g., panretinal photocoagulation) or non-existent for RP (Campochiaro et al., 2015; Kern et al., 2010; Punzo et al., 2012). Rod cell-based oxidative stress has been implicated as a common pathogenic event underlying eventual histopathology in DR and RP, among other retinopathies (Fig. 1) (Cingolani et al., 2006; Dong et al., 2006; Du et al., 2013; Kassen et al., 2009; Komeima et al., 2006; Rogers et al., 2007; Shen et al., 2005; Usui et al., 2009b). A likely source of rod cell oxidative stress is the ellipsoid region of the inner segment which contain ~75% of retinal mitochondria (Johnson, Jr. et al., 2007; Medrano and Fox, 1995; Perkins et al., 2003). Notably, other abnormalities in, for example, DR, such as inflammation, appear as downstream consequences of oxidative stress (Du et al., 2013). In addition, the identification of rod cells as the main contributor to retinal oxidative stress complicates interpretation of systemic treatments that were previously thought to correct primarily endothelial cell oxidative stress.

Fig. 1.

Simplified representation of key events in DR and RP over time; DR and RP time courses are not to scale and are not comparable. The key point here is that rod cell metabolic oxidative stress and dysfunction – either from a metabolic disturbance (DR) or genetic abnormality (RP) – are important early events.

These considerations provide a rationale for focusing on the evaluation of anti-oxidant therapy on rod cells in vivo. Much better outcomes in DR and RP can thus be expected if antioxidant therapy is started before gross changes are evident. However, testing the effectiveness of an antioxidant against rod oxidative stress has mostly relied on post-mortem studies and/or observing improvements in animal models following (often pleiotropic) anti-oxidant therapy (Berkowitz et al., 2007c, 2012a, 2009b; Campochiaro et al., 2015; Du et al., 2013, 2015; Fukuda et al., 2014; Galbinur et al., 2009; Jaliffa et al., 2009; Komeima et al., 2006; Rohrer et al., 2004; Sanz et al., 2007; Usui et al., 2009a; Yang et al., 2007; Zeng et al., 2014; Zheng et al., 2007). Such approaches i) are often unable to determine whether oxidative stress in rod cells per se has been corrected, and ii) are not useful for examining the same animal over time or for examining rod cell oxidative stress in patients. A need remains to non-invasively measure the efficacy of antioxidant therapy against early rod cell compartment-specific oxidative stress in diseases like DR and RP in order to personalize anti-oxidant therapy with regard to timing and dosing in vivo. In vulnerable neurons, including rod cells, one of the first consequences of pathogenic oxidative stress is cell dysfunction [e.g., (Du et al., 2013, 2015; Mao et al., 2014; Roddy et al., 2012; Wang and Michaelis, 2010)] (Fig. 1). Thus, we have focused this review on efforts on developing an optimized imaging platform that can non-invasively measure the effectiveness of antioxidant treatment in correcting rod cell compartment-specific pathophysiology before the appearance of other clinical biomarkers in disease.

1.1. Problems with current technology

Conventional non-invasive approaches measure either blood vessel/flow (optical coherence tomograph (OCT) angiography, fluorescein angiography), anatomy (fundus photography, OCT, adaptive optics), or a global function (ERG). Behavioral tests (e.g., optokinetic tracking) examine whole-animal visual performance (e.g., acuity and contrast sensitivity) usually under phototopic conditions. Thus, these common approaches do not measure the package of rod-specific functions together with retinal vascular and choroid functions that are specifically evaluated by MRI (see below).

At present, OCT is the gold-standard for evaluating anatomical changes in the laminar structure of the retina in vivo but provides little functional information especially about rod cells. The most common method for evaluating rod photoreceptor function in vivo is the electroretinogram (ERG), which measures an integrated panretinal signal. Several studies have used ERG used to evaluate anti-oxidant therapies on diabetes-induced rod dysfunction, however, ERG reports on the entire retina and thus provides no panretinal spatial resolution (Barile et al., 2005; Horio et al., 2004; Johnsen-Soriano et al., 2008; Midena et al., 1989; Samuels et al., 2012). Multi-focal ERG (mfERG) can distinguish electrical responses from different regions panretinally but suffers from extensive light scattering in small rodent eyes (Ball and Petry, 2000; Nusinowitz et al., 1999). In other words, electrophysiology is a relatively insensitive tool for evaluating focal dysfunction in common preclinical models. ERG measures an important but limited aspect of global rod function: movement of monovalent ions on a millisecond time scale after a flash of light in vivo. However, ERG does not report on changes in the flux of important divalent ions such as calcium. It is likely that this is why ERG has not been able to predict loss of visual performance in models of retinal degeneration and in aging in the absence of overt pathology (Bissig et al., 2013; McGill et al., 2012). ERG thus does not capture the increased calcium channel function that was found to be highly predictive of later declines in visual performance (Bissig et al., 2013). These considerations directly speak to the need for MRI methods that do quantitatively evaluate calcium channel function in vivo with high spatial resolution.

In addition, each approach is incomplete on its own (e.g., ERG provides function information without spatial resolution, OCT provides high spatial resolution but not information about rod function). Thus, the information available from each approach cannot be integrated into a coherent whole. For these reasons, the above methods have had limited success in evaluating the efficacy of antioxidant treatment, specifically on rod cells early in the course of disease, before the appearance of histopathology. The goal of this review is to highlight that MRI fills a non-invasive technology need not addressed by conventional technologies for i) a package of quantitative and non-invasive measurement of rod cell compartment-specific function together with retinal vessel and choroidal functions in vivo, ii) integrating retinal structure and rod function into a coherent picture and iii) early detection and evaluation of anti-oxidant treatment efficacy before the gross appearance of disease.

2. Previously learned lessons

Here, we briefly summarize three examples that illustrate how earlier and lower spatial resolution MRI studies of retinal disease and its treatment showed alternate views of retinal disease, and came to alternate conclusions that need to be evaluated alongside that of other studies (Berkowitz and Roberts, 2008, 2010; Trick and Berkowitz, 2005).

Example 1:

Tissue ablation surgery (e.g., panretinal photocoagulation) is the standard of care for proliferative and non-proliferative retinopathy. This surgery is commonly assumed to relieve retinal hypoxia at the border of vascular and avascular retina that is driving abnormal new vessel growth (i.e., retinal neovascularization). If this assumption is true, retinal hypoxia should precede the appearance of neovascularization. However, magnetic resonance studies did not find evidence to support a special early role of retinal hypoxia in proliferative retinopathy models, a conclusion in-line with that of other labs (Alder et al., 1991; Berkowitz, 2008; Diederen et al., 2006; Lau and Linsenmeier, 2014; Ola et al., 2006; Reeves et al., 2002; Wanek et al., 2014; Wright et al., 2011; Zhang et al., 2003).

In non-proliferative diabetic retinopathy models, the evidence to-date has failed to support a role for hypoxia at least before the appearance of histopathology (Alder et al., 1991; Diederen et al., 2006; Lau and Linsenmeier, 2014; Ola et al., 2006; Reeves et al., 2002; Wanek et al., 2014; Wright et al., 2011). To-date, only one study of three cats diabetic for 10 years has reported evidence for hypoxia together with vascular histopathology (Linsenmeier et al., 1998); it is not known if this is also true in more common animal models, such as the diabetic rat. Here, we present new data that begins to address this question. Pre-retinal oxygen tension was measured by a magnetic resonance method after 9–10 mo of diabetes, a time when vascular lesions are present in a non-proliferative diabetic rat model. In control and 9–10 mo streptozotocin-induced diabetic male Sprague Dawley rats, preretinal oxygen levels were measured with 19F NMR of a preretinal vitreous perfluorocarbon droplet during room air or 12% oxygen breathing in vivo (Arteel et al., 1998; Barber et al., 2005; Feit-Leichman et al., 2005; Handa et al., 1996; Kern et al., 2000; Kowluru et al., 2001; Zhang et al., 2003); the 12% condition was included to demonstrate the sensitivity of the method to hypoxia. Because of the small distances, a pre-retinal oxygen measurement reflects to some degree inner retinal oxygen levels but not as sensitively as an intraretinal oxygen measurement (Arteel et al., 1998; Barber et al., 2005; Feit-Leichman et al., 2005; Handa et al., 1996; Kern et al., 2000; Kowluru et al., 2001; Zhang et al., 2003). A level of hyperglycemia of 400–550 mg/dl was maintained for the duration of diabetes. Core temperature and blood gas parameters measured after the experiment and during either 12% oxygen breathing or room air were appropriate for each challenge (data not shown). In control rats, no significant difference (P > 0.05) was found between pre-retinal PO₂ during room air in the 3 and 9–10 mo control groups (32 mm Hg ± 4 mm Hg, n = 6; 23 ± 2 mm Hg, n = 7, respectively; mean ± SEM). Pre-retinal vitreous oxygen tension during 12% oxygen inhalation (1 ± 4 mm Hg, n = 6) was significantly (P < 0.05) lower than that measured during room air breathing (~21%; 32 mm Hg ± 4 mm Hg, n = 6). Pre-retinal vitreous oxygen levels during room air breathing were not statistically different between 9 and 10 mo controls (23 ± 2 mm Hg, n = 10) and rats diabetic for 10–13 months (25 ± 8 mm Hg, n = 3) While these new data do not completely rule out the possibility of small areas of focal or subclinical hypoxia in retinas, it is hard to rationalize how such changes could play a major role in the development of a number of changes found in the retina during experimental diabetes, including upregulation of the potent mitogen vascular endothelial growth factor (VEGF) (Frank, 2004), and apoptosis of retinal cells (Bussink et al., 2003; Kern et al., 2000). These results, limited by measuring from a single location in the pre-retinal vitreous, do not support the presence of retinal hypoxia during the appearance of clinically relevant retinal vessel histopathology in the rat model. Overall, little evidence is available to support an early (i.e., pre-histopathologic) role of hypoxia in proliferative and non-proliferative retinopathy, although the role of hypoxia later in the disease remains unclear. More work is needed to identify early, non-hypoxia-based mechanisms in order to improve upon current treatment strategies.

Example 2:

It is a common belief that a leaky blood retinal barrier is the primary cause of diabetic retinal edema and subsequent vision loss. By taking advantage of MRI inherent sensitivity to water spin density and self-diffusion, together with MRI measurements of retinal thickness, it was established that diabetic male Sprague Dawley rats present with retinal edema from 2 to 9 mo of diabetes; 2 mo was the earliest time point examined by MRI and others find evidence for increased retinal thickness as early as 2 weeks of diabetes in the male Sprague Dawley rat (Berkowitz et al., 2012b; Clermont et al., 2011). However, there is less agreement as to when the blood retinal barrier (BRB) is disrupted. Studies find evidence for increased leakiness starting as early as 1 weeks of diabetes or as late as 8 mo of diabetes (Berkowitz et al., 2004; Xu et al., 2001). The reasons for the differences in the appearance of BRB damage in diabetic models are unclear. We note that many of the studies reporting earlier loss of BRB integrity involve tracers such as Evans blue and fluorescein that bind to albumin (Antonetti et al., 1998; Xu et al., 2001). In contrast, studies that report later BRB breakdown involve strictly passive tracers such as Gd-DTPA and sucrose (Berkowitz et al., 2004; Ennis and Betz, 1986). A potential unifying explanation is that the different tracers report on different types of BRB damage over time with earlier injury occurring to albumin-based paracellular transport mechanisms and later leakage due to loss of tight cell–cell junctions (Berkowitz et al., 2012b, 2004; Ennis and Betz, 1986; Grimes, 1985, 1988). In either case, a common misperception is that tracer accumulation equates to increased permeability. It is more accurate to view tracer accumulation as a function of tracer influx, the tracer plasma concentration integral over time, and tracer efflux (Berkowitz et al., 1992). Thus, if influx of a tracer did not change but tracer efflux decreased, one would measure net accumulation of the tracer. Indeed, new evidence supports the view that water efflux via Mueller cells, for example, is impaired by diabetes (Reichenbach et al., 2007). Growing evidence is expected to redirect efforts away from studies of water influx pathways as a cause of diabetic retinal edema to an improved appreciation of cellular mechanisms underlying impaired water removal.

Example 3:

In proliferative and non-proliferative disease, the most commonly detected abnormality across many labs and methods (including MRI) is retinal vessel dysfunction during, for example, a hyperoxic provocation (i.e., a stress test) (Kern et al., 2010; Lorenzi et al., 2010; Trick and Berkowitz, 2005). Importantly, treatments that correct retinal vessel dysfunction also correct later histopathology (Berkowitz et al., 2013; Trick and Berkowitz, 2005). Correction of impaired autoregulation early in the course of the disease as measured by MRI, acts as a prognostic treatment efficacy biomarker (Berkowitz et al., 2013; Berkowitz and Roberts, 2008, 2010; Trick and Berkowitz, 2005). From these results, evaluating retinal autoregulation using non-MRI methods that measure, for example, retinal vessel diameter, may be a useful approach in the clinical management of treatment efficacy in DR.

These three examples underscore MRI as a valuable discovery tool in the study of retinal pathophysiology and for gauging early treatment efficacy in vivo. However, in those early MRI studies, resolution was too low to interrogate rod cell function.

3. Why not use fMRI to study rod activity?

Common MRI methods for evaluating neuronal function involve approaches such as blood-oxygen level dependent (BOLD) or arterial spin labeling (ASL) approaches that capitalize on task-dependent hemodynamic changes to spotlight regions of the brain that are active (Detre and Wang, 2002). These methods are broadly used, in part, because they are reliable and do not involve injecting a contrast agent (Kim and Ogawa, 2012). BOLD, ASL, and related methods, have been modified for use to study the retina but at low resolution relative to the retinal thickness (Duong et al., 2002; La Garza et al., 2011; Li et al., 2008; Muir and Duong, 2011; Nair et al., 2011; Shih et al., 2011; 2012). Critically, BOLD and ASL only measure hemodynamically-dependent indices linked with neuronal activity and rod cells exist in an avascular environment. Thus, fMRI approaches are not expected to be particularly useful for specifically interrogating rod cell function.

4. Breakthrough

In 2006, the first images of rat retina in vivo were published with a remarkable axial spatial resolution of 23.4 μm (Luan et al., 2006). At the time each retinal image collection required about 1 h. Over the years, with better MRI systems and acquisition schemes, images with this resolution can be collected in just a few minutes; similarly high resolution images can also be obtained in other common laboratory animals, for example in mice and zebrafish (in collaboration with Dr. Ryan Thummel, Wayne State University) (Fig. 2) (Chan et al., 2012; Chen et al., 2008, 2011; Duong, 2014; Wang et al., 2011).

Fig. 2.

Representative MRI images (axial resolution 23.4 μm) of three common laboratory models acquired using established protocols (Berkowitz et al., 2014; Bissig et al., 2013); the images are proportionately sized based on ocular anatomy.

This 23.4 μm resolution is important because it allows for the collection of MRI data that can distinguish inner retina structure/function from that in the outer retina aided by the 200 μm thick retina having a well-defined laminar structure (Berkowitz et al., 2009a, 2010, 2012a, 2006, 2007d; Bissig and Berkowitz, 2011). The inner retina has a multi-laminar vasculature and a large number of different and overlapping cell types. These features can confound interpretation of MRI data from the inner retina. On the other hand, MRI of the outer retina is more readily interpreted because the outer retina is avascular and, in mice, ~97% of the cells are rods (Carter-Dawson et al., 1978). In addition, advantage of the distinct rod cell compartment-specific functions performed by rod nuclear and outer segment layers allows further spatial clarity (Fig. 3). These functions include i) activity-dependent voltage-gated L-type calcium channels (LTCCs) in the outer nuclear layer (ONL) and inner segments (IS) but not in the outer segments (OS), ii) light-dependent formation of tetrameric arrestin1 in the ONL and IS and its translocation as a monomer to the OS, and iii) light-evoked expansion of the extracellular space around the OS (i.e., the subretinal space, SRS) (Fig. 3) (Berkowitz et al., 2015a; Bissig and Berkowitz, 2012). Thus, MRI evaluates the same set of rod cells for compartment-specific function in vivo using different types of acquisitions (see below for details; summarized in Fig. 4) (Berkowitz et al., 2013, 2015a, 2015c; Bissig and Berkowitz, 2012). In addition, four “free” metrics are also obtained: retinal and choroidal thickness, and light-dependent expansion of the inner retinal circulation and choroid (Fig. 4). Further confidence is provided by the spatial agreement between transretinal MRI maps of function and OCT (Fig. 4).

Fig. 3.

Simplified illustration of the rod cell functionome (white boxes in center) in dark (left) and light (right) as assessed by multi-functional MRI metrics. The array of essential information measured by MRI (shown in the white boxes) is unavailable from other imaging methods. Inner retinal circulatory plexi are not shown. Drawing courtesy of Jawan Gorgris.

Fig. 4.

Summary of multi-functional MRI (graphs) spatially registered to OCT (top image, left). Rod functions in dark (black line) vs. light (pink line) in non-diabetic (left column) and diabetic (right column) mice. In summary, diabetes impairs light-evoked functions (except arrestin1 and its translocation) and these abnormalities were correctable with antioxidant treatments (Table 1). N’s for the dark and light data, respectively, are as follows: LTCC graphs (wt: 19 and 11, diabetic: 6 and 7); Sub-retina space graphs: (wt: 23 and 23, diabetic: 9 and 9); 1/T1ρ graphs (wt: 7 and 7, diabetic: 5 and 5). Light-evoked choroidal expansion is maintained in diabetics but to a significantly reduced degree (bargraph, paired dark/light data, n = 18 (wt) and 9 (diabetic)); an antioxidant did not correct this abnormality. *, P < 0.05. Data for each graph is obtained from central (+/− 0.4 − 1 mm from the optic nerve head) retina and shown as a function of distance from retina/non-retina borders, where 0% is the vitreous/retina border and 100% is the retina/choroid border. Above graphs: simplified schematic of retina and support circulations (Paques et al., 2003;Wangsa-Wirawan and Linsenmeier, 2003; Zhi et al., 2014). Red horizontal bracket and star above profiles indicates specific retinal regions with significant (P < 0.05) light-evoked differences. Based on OCT data (Fig. 4), vertical dashed lines approximate layer locations in retina.

In this review, we cover material from 2006 to the present, focusing on the use of MRI to simultaneously measure the above functions of rod cell compartments that are not effectively evaluated by other methods in vivo.

5. Rod nucleus/synapse/inner segment LTCCs

LTCCs are the major calcium entry path into rod cells (and other retinal neurons), and play an essential role in rod function (Chen et al., 2014; Krizaj, 2012; Molnar et al., 2012). For example, in the dark, rod cell membranes are depolarized resulting in the sustained opening of synaptic LTCCs (Fig. 5). Persistent opening of these LTCCs triggers continuous release of the neurotransmitter glutamate (Schmitz and Witkovsky, 1997). Thus, LTCCs bridge the two essential functions of the rod cell, light detection and signaling to the rest of the retina and brain; the role of LTCC in the other compartments of the rod are less clear (Xiao et al., 2007). Rod compartment-specific LTCC function in vivo can be evaluated using high resolution manganese-enhanced MRI (MEMRI).

Fig. 5.

Cartoon summarizing impact of rod cell membrane hyperpolarization in the light (left) and depolarization in the dark (right) on influx of manganese.

5.1. Brief history of MEMRI

In 1997, Alan Koretsky and his group demonstrated the potential value of using a non-toxic dose of the paramagnetic ion manganese as an intracellular MRI contrast agent whose accumulation occurred primarily in active neurons via voltage-gated LTCCs (Berkowitz et al., 2007b, 2009a, 2011, 2012a, 2013; Carlson et al., 1994; Cross et al., 2007; Drapeau and Nachshen, 1984; Jacob, 1990; Lin and Koretsky, 1997; Tofts et al., 2010). LTCCs play essential roles modulating neuronal activity including neurotransmitter release, and have a privileged role in activity-dependent gene expression via the transcription factor cAMP response element-binding protein (Mermelstein et al., 2000; Schmitz and Witkovsky, 1997). Manganese has three desirable properties. First, it enters cells over several hours and leaves slowly (days to weeks depending on the tissue) so interpretation is not confounded by efflux or time concerns (Tofts et al., 2010). Second, activity-dependent manganese entry into cells occurs primarily via LTCCs (Berkowitz et al., 2007b, 2009a, 2011, 2012a, 2013; Carlson et al., 1994; Cross et al., 2007; Drapeau and Nachshen, 1984). Importantly, based on the absence of toxicity effects at the present dose of manganese, there is little evidence to suggest that manganese blocks the calcium channels and prevents calcium entry (for more detail see section 9) (Berkowitz et al., 2013). Third, there is no endogenous agonist-releasable manganese store (Jacob, 1990), so the appearance of retained manganese unambiguously signals that its entry into the cytoplasm originated from outside the cell and not from an internal store (Jacob, 1990). For these reasons, MEMRI is a powerful discovery tool that is also useful for translating electrophysiology results obtained in isolated cells to preclinical models, and eventually, into the clinic (Tofts et al., 2010). Spatial agreement in activated regions between MEMRI and BOLD was noted in control animals (Duong et al., 2000). The Koretsky approach was, however, limited by the need to transiently breakdown the blood brain barrier in an anesthetized animal during a functional task. Breaking down the blood brain barrier even transiently is physiologically problematic. Nonetheless, it was clear from these early studies that MEMRI generates functional maps at much higher spatial resolution than other fMRI methods, as well as provides unique and direct interrogation of the extent of neuronal activation not measured by other fMRI methods.

In 2005, Dan Turnbull and his group reported that there was not a need to breakdown the blood barrier to encode function as long as the functional task lasted for several hours after systemic injection of manganese (Yu et al., 2005). The Turnbull approach offered another major advantage: functional encoding outside of the magnet while the animal is awake and freely moving. Only after the prolonged functional task is presented for a fixed period of time is the animal anesthetized and imaged to measure how much manganese accumulated. This represented an opportunity for a new approach to functional retinal imaging: combining Turnbulls’ prolonged encoding of functional information with very high resolution images of the retina. MEMRI has been called the imaging modality of choice for non-invasively studying retinal ion homeostasis (Ramos de Carvalho et al., 2014). Much of the basic operational aspect of the MEMRI has been reviewed previously (Berkowitz et al., 2013).

Below, the sensitivity and physiologic accuracy of MEMRI for studying rod cell stimulation is demonstrated by comparing the extent of manganese accumulation in different rod compartments to known modulators of rod function, such as light-evoked changes in cGMP-gated channels, the visual cycle, and horizontal cell (HC) inhibitory signaling (Berkowitz et al., 2009a, 2006).

5.2. cGMP and the visual cycle

In the dark, rod cell membranes are persistently depolarized by open cGMP channels resulting in prolonged opening of the associated LTCCs. In line with this, MEMRI measures a greater uptake of manganese in rod nuclear and inner segment layers in dark vs. light; further, the uptake was linearly and inversely proportional to the log of light intensity used for adaptation, similar to that reported between background light adaptation and extracellular current in isolated rod photoreceptors (Fig. 6) (Berkowitz et al., 2009a, 2006). Genetic inhibition of just rod phototransduction in rod transducin knockout mice (GNAT1^−/−) prevents both cGMP opening and the differential uptake of manganese in rod nuclear and inner segment layers between dark and light adapted mice (Berkowitz et al., 2014). Further, visual cycle impairment at the level of RPE65 reduces 11-cis-retinal production, leading to a buildup of unbound opsin with closure of cGMP channels and LTCCs (Berkowitz et al., 2009a). Consistent with this expectation, pharmacologic suppression of RPE65 resulted in a decrease in the amount of manganese taken up in the outer retina relative to that in dark-adapted mice without such visual cycle manipulation (Berkowitz et al., 2009a). These data were used to support MEMRI’s sensitivity to impaired visual cycle.

Fig. 6.

Functional LTCC topography as measured by MEMRI in dark and light adaptation. (A) Data for C57Bl/6 mice (wt); graph and graphing conventions are as in Fig. 4. B) and C) illustrate how manganese uptake in inner retina (0–50% depth) or outer retina (50–100% depth), as measured using T1 weighted MRI, response to changes in light intensity, respectively, in male albino control Sprague—Dawley rats after exposure to the following light intensities: 1.8 ± 0.7 (n = 10, mean ± SEM), 51.3 ± 11.7 (n = 5), and 250.2 ± 19.3 (n = 6) lux. (Berkowitz et al., 2009a). In C), the red line indicates a significant (P < 0.05) linear correlation. *, P < 0.05.

Considerably lower than normal uptake of manganese was measured transretinally in 6 week old rd12 mice, which have a loss-of-function mutation in RPE65; notably, this decrease was correctable with systemic 11-cis-retinal treatment (Berkowitz et al., 2009a; Pang et al., 2005). However, more recent research showed that rd12 mice undergo a significant loss of cones by 5 weeks of age, potentially explaining the impact of this mutation across the whole retina (Fig. 7); in addition, 11-cis-retinal can exert anti-oxidant effects separate from the visual cycle, thus providing an alternative benefit, as measured by MEMRI (manuscript in press) (Pang et al., 2010).

Fig. 7.

Functional LTCC topography as measured by MEMRI in dark adapted wt mice (n = 6) or 6 week old rd12 mice (n = 7); graphing conventions are as in Fig. 4. *, P < 0.05.

It is worth noting while different groups agree that the outer retina of rats accumulates more manganese in the dark than light adapted conditions, with this uptake sensitive to diabetes (Berkowitz et al., 2009a, 2006; De La Garza et al., 2012; Muir et al., 2015), there is an apparent discrepancy regarding uptake manganese uptake in the inner retina with dark and light. For example data illustrated in Fig. 4 finds no differences in the extent of manganese uptake in the inner retina between dark and light; however Duong et al. reported higher MEMRI activity in light compared with dark adaption (De La Garza et al., 2012; Muir et al., 2015). There are several possible explanations for the different results in the inner retina measured by the two groups. One is that we administer manganese to awake and freely moving animals, whereas Duong et al. injects anesthetized animals that are then allowed to wake up during dark or light adaptation; this experimental design raises the possibility of a residual anesthesia effect on retinal manganese uptake. Perhaps the most likely difference is that one group generally performs its studies in the absence of patching, and the other patches one eye to obtain dark adapted and light adapted retinae in the same animal. In an earlier paper, a significant light-dependent Mn²⁺ uptake in the inner retina was noted, perhaps involving a pathway from the patched eye through the brain to the unpatched eye (Bissig and Berkowitz, 2011). These considerations do not support a discrepancy between groups, but do highlight the sensitivity of MEMRI outcomes to the experimental design.

These results establish the sensitivity and accuracy of MEMRI to light detection as it impacts rod cell LTCCs.

5.3. Inhibitory signaling and rod cells

In 2007, sodium iodate (24 h post injection of 30 mg/kg, IP), an insult that can cause RPE damage, leading to blood retinal barrier breakdown, photoreceptor degeneration, retinal dysfunction, and loss of visual behavior was observed to also produce a greater than normal uptake of manganese in rods (and cells of the inner retina) (Berkowitz et al., 2007d). Notably, the supernormal uptake was not linked with iodate-induced opening of the blood retinal barrier (Berkowitz et al., 2007d). The interpretation of these results was not clear but it has been suggested that iodate-induced damage to the photoreceptors produces a loss of glutamatergic signaling to horizontal cells (HC) which in turn impaired HC inhibitory signaling back onto the rods (Schubert et al., 2006). Normally, HC depolarization in the dark following glutamatergic signaling results in closure of LTCCs on rods (Babai and Thoreson, 2009; Liu et al., 2013). The purpose of HC inhibitory signaling in bright light is most commonly understood in terms of i) controlling conegenerated synaptic gain and ii) lateral inhibition onto cone cells forming the basis of the center-surround organization of vision (Thoreson and Mangel, 2012; Yang and Wu, 1991). In dim light, HC inhibitory signaling onto rod cells has also been documented and suggested to fulfill these same two purposes (Babai and Thoreson, 2009; Mansergh et al., 2005; Morgans et al., 2001; Schmitz and Witkovsky, 1997). In total darkness, HC inhibitory signaling may act to throttle down the extent of LTCC opening to regulate how much calcium enters into rod (and possibly cells of the inner retina). It was beyond the scope of this review to investigate the exact mechanism(s) by which HCs communicate with photoreceptors, a currently active area of research by other labs, or with inner retina (Kamermans et al., 2001; Thoreson and Mangel, 2012; Yang and Wu, 1991).

When HC inhibitory signaling to rod cells is impaired, the extent of LTCC opening is expected to be greater than normal, allowing more calcium and manganese into rod cells (and possibly cells of the inner retina). HC inhibitory signaling is involved in the processing of light information by i) control of photoreceptor-generated synaptic gain and ii) lateral inhibition onto photoreceptor cells forming the basis of the center-surround organization of vision (Thoreson and Mangel, 2012; Yang and Wu, 1991). HC inhibitory signaling may also occur to the inner nuclear layer but that has been less studied (Thoreson and Mangel, 2012; Yang and Wu, 1991).

Thus, a recent study investigated whether loss of glutamatergic signaling to HC leads to impaired HC inhibitory signaling back onto the rods in the dark using MEMRI and mice null for proteins known to regulate glutamatergic-evoked HC inhibitory signaling to photoreceptors (i.e., Ca_v1.4 and arrestin-1) in the light (Babai and Thoreson, 2009; Berkowitz et al., 2015b; Mansergh et al., 2005; Morgans et al., 2001; Schmitz and Witkovsky, 1997). The data from the Ca_v1.4 experiments are summarized in Fig. 8 and supports the idea that impaired photoreceptor glutamateric signaling to the HC (as would also be expected following sodium iodate treatment) results in loss of HC inhibitory signaling back to the photoreceptors leading to supernormal uptake of manganese in both rods and cells of the inner retina (Berkowitz et al., 2015b). The above data provide the first application of MEMRI to study inhibitory signaling in vivo.

Fig. 8.

Functional LTCC topography as measured by MEMRI in dark adapted wt mice (closed black circles, n = 19), Ca_v1.4^−/− mice (closed gold circle, n = 5), and cx57^−/− mice (closed blue circle, n = 5). Graphing conventions are as in Fig. 4 (Berkowitz et al., 2015b). *, P < 0.05.

A recent study in zebrafish suggested the controversial hypothesis that HC-specific connexin 57 (Cx57) regulates inhibitory feedback to photoreceptors (Ciolofan et al., 2007; Dedek et al., 2008; Hombach et al., 2004; Kamermans et al., 2001; Klaassen et al., 2011; O’Brien et al., 2012a,b; Pandarinath et al., 2010; Shelley et al., 2006). Connexins, transmembrane proteins that form channels that make up hexameric hemichannels between apposed cells, provide a communication path between neurons and contribute to information processing. The availability of connexin57 knockout mice (cx57^−/−; a kind gift from Dr. Sheila Nirenberg, Weill Medical College of Cornell University) provided an opportunity to test this hypothesis with MEMRI. Here, we present new data that are consistent with the results in zebrafish in that cx57^−/−mice demonstrate a greater than normal uptake in their rod cells (Fig. 8). In addition, a greater than normal uptake in the inner retina was noted in the cx57^−/− mice, providing provisional evidence for HC inhibitory “feed forward” signaling (Thoreson and Mangel, 2012; Yang and Wu, 1991). Together, this supernormal and symmetric uptake phenotype at least raises the possibility for a role for cx57-modulated HC inhibitory signaling in the dark adapted awake and freely moving mouse in vivo.

Another gap junction protein at the photoreceptor synapse, but not known to be involved in HC inhibitory signaling, is the neuronal gap junction connexin 36 (Cx36), which is also expressed in cones, particularly at the pedicle and cone-to-rod contacts, but not in rods per se; Cx36 is also expressed in retinal amacrine and ganglion cells (Feigenspan et al., 2004; Güldenagel et al., 2001; O’Brien et al., 2012a, b; Schubert et al., 2005). Functionally, Cx36 is thought to contribute to regulation of cone-rod interactions and rod synaptic transmission based on post-mortem examination (Asteriti et al., 2014; Güldenagel et al., 2001). Cx36 is expressed at the outer plexiform layer in a laminar and non-interacting arrangement with cx57 (Ciolofan et al., 2007; Deans and Paul, 2001; O’Brien et al., 2012a, b). Here, new data are presented of dark adapted cx36^−/− mice (a kind gift from Dr. Marla Feller) demonstrating focal regions of subnormal uptake, consistent with a role for Cx36 in the inner and outer segments, and in the inner plexiform and ganglion cells layers (Fig. 9). These MEMRI data support independent contributions from cx57 and cx36 in the dark adapted mouse in vivo (Ciolofan et al., 2007).

Fig. 9.

Functional LTCC topography as measured by MEMRI in dark adapted wt mice (n = 19) or cx36^−/− mice (n = 5); graphing conventions are as in Fig. 4. *, P < 0.05.

Together, the data establish MEMRI as the first objectively useful method for evaluating light detection and processing as it impacts rod cell LTCCs in vivo.

5.4. MEMRI and Cav1.* subtypes

The retina does not contain a single type of LTCC but instead three LTCC subtypes, Ca_v1.2, 1.3, and 1.4 channels, each with different roles based on their different biophysical properties. Arguably the most important subtype is the Ca_v1.4 channel located only at the photoreceptor synapse. Ca_v1.4 channels play an essential role in vision since Ca_v1.4^−/− mice are blind and mutations in the CACNA1F gene, which encode Ca_v1.4 LTCC α1 subunits, are linked with loss of vision in Åland Island Eye Disease, cone-rod dystrophy, X-linked retinal disorder, night blindness-associated transient tonic down-gaze, and incomplete congenital stationary night blindness (Baumann et al., 2004; Burtscher et al., 2014; Busquet et al., 2010; Koschak et al., 2003; Mansergh et al., 2005; Morgans et al., 2001; Strom et al., 1998; Xiao et al., 2007). In contrast, the functional role of Ca_v1.3 channels is less clear, in part because i) current antibodies for identifying these channels and used for immunological localization are non-specific and ii) a selective antagonist has not yet been established (Zou et al., 2012). Ca_v1.3 channels have been documented in cells of the inner retina and in retinal pigment epithelium cells based on electrophysiological evidence, and their mRNA exists in rod cells and cells of the inner retina (Habermann et al., 2003; Morgans et al., 2005; Rosenthal et al., 2001; Xiao et al., 2007; Xu et al., 2002; Zou et al., 2012). However, Ca_v1.3^−/− mice do not exhibit an impaired visual phenotype as measured by electrophysiology and water maze testing (McKinney and Murphy, 2006; Wu et al., 2007). The remaining major LTCC subtype, Ca_v1.2, is expressed in the inner plexiform and ganglion layers, but not rod cells in young mice, although in older mice Ca_v1.2 LTCCs appear to be functional in the outer retina in vivo (Berkowitz et al., 2014; Busquet et al., 2010).

To separately interrogate LTCC Ca_v1.* subtypes, mice were pretreated (i.e., before the injection of manganese) with D-cis-diltiazem (DIL) at a dose that appeared to antagonizes only Ca_v1.2 channels in vivo (Berkowitz et al., 2014). DIL IC50’s, generated in different labs using expressed channels in cells, suggest some subtype specificity [i.e., IC50 ~45 μM (Ca_v1.2) vs. ~326 μM (Ca_v1.3) and ~92 μM (Ca_v1.4)] (Baumann et al., 2004; Berkowitz et al., 2014; Bissig et al., 2013; Glossmann et al., 1983; Tarabova et al., 2007). However, the exact DIL IC50’s for the LTCC subtypes in the retina in vivo are not known (Berkowitz et al., 2014; Bissig et al., 2013). Nonetheless, spatial agreement was noted between where Ca_v1.2 is located on immunohistochemistry (inner retina) and DIL-induced focal suppression of manganese uptake (inner retina) (Baumann et al., 2004; Berkowitz et al., 2014; Bissig et al., 2013; Glossmann et al., 1983; Tarabova et al., 2007). Ca_v1.3, an order of magnitude less sensitive than Ca_v1.2 channels to DIL based on the above IC50’s, is thought to be more widely expressed than just inner retina and evidence for DIL-related suppression in other parts of the retina of wildtype (wt) mice was noted (Xiao et al., 2007). Together, these data in vivo are consistent with the notion that Ca_v1.3 channels are less responsive to DIL treatment than Ca_v1.2 channels (Lipscombe et al., 2004; Xu and Lipscombe, 2001). If DIL substantially antagonized Ca_v1.4 channels, one expects a reduction in HC inhibitory signaling to the rod cells (see above) and thus a greater than normal uptake of manganese. A supernormal uptake in rod cells was not seen in the mice treated with DIL (Berkowitz et al., 2014). Collectively, these considerations support the use of DIL for unraveling the contribution of LTCC Ca_v1.* subtypes in vivo using MEMRI (Bissig et al., 2013). Thus, MEMRI alone is not able to separate the functional contributions from cav1.* channels. However, MEMRI with and without DIL treatment, together with the highly localized Cav1.4 at the photoreceptor synapse, laminar-structure of the retina, and high spatial resolution of MRI has been useful in unraveling the various subtype contributions in vivo.

MEMRI with and without DIL studies have raised the possibility that the number and/or spatial location of LTCC subtypes varies within the retina in, for example, age and disease; a similar phenomenon has been documented in the brain (Jurkovicova-Tarabova et al., 2012; Sinnegger-Brauns et al., 2009). For example, we find evidence that manganese uptake in rod cells and other cells of the retina steadily increases with age (Fig. 10). MEMRI with and without DIL experiments suggest that this was due to increased Ca_v1.3 expression that was predictive of visual performance declines in healthy aging (Bissig et al., 2013); exactly how increased Ca_v1.3 leads to vision loss remains unclear. Also, in Ca_v1.3^−/− mice and in 10 mo old mice, DIL suppressed uptake in the outer plexiform layer and outer nuclear layer, suggesting compensatory up regulation of retinal Ca_v1.2 channels not present in wt or in younger mice, respectively; knocking out specific proteins regulating photoreceptor synapse activity also modified the array of Ca_v1.2 channels in the retina as measured by MEMRI with and without DIL (Jurkovicova-Tarabova et al., 2012). Intriguingly, discordant reports of the efficacy of the Ca_v1.2-specific antagonist DIL in RP models may be, at least in part, a result of this plasticity of the Cav1.2 subtype (Barabas et al., 2010; Berkowitz et al., 2011, 2014; Bissig et al., 2013).

Fig. 10.

Representative maps of tissue R₁ (in units of s⁻¹) from baseline (no Mn²⁺, left half of each composite image) and Mn²⁺-injected long Evans rats (right half of each composite image) (Bissig et al., 2013). In Mn²⁺-injected rats, R₁ — which is linearly related to tissue Mn²⁺ concentration (Chuang et al., 2009) — increases with age; this change was not linked with blood–retinal barrier breakdown (Bissig et al., 2013). In addition, the no-manganese baseline R₁s were stable over time (Bissig et al., 2013).

Mapping the number and spatial location of retinal LTCC subtypes in vivo using MEMRI with and without DIL will likely help guide/optimize calcium channel therapy in aging and in disease (Fox et al., 2003).

5.5. Other photosensitive cells

Rod cells are not the only photosensitive cells that can be studied by MEMRI. For example, MEMRI readily detects light-stimulated changes in intrinsically photosensitive retinal ganglion cells in postnatal day 7 (P7) mice (when ipRGC content is high and photoreceptors are not functional), and in inner retinal cells genetically modified to express channelrhodopsin in rd1 mice (optogenetics) (Berkowitz et al., 2010; Ivanova et al., 2010). In both cases, MEMRI outcomes matched the known physiology of these channels which function in an opposite manner to that of rod cells by having greater opening in the light than in the dark. It remains unclear at present whether the manganese entered the cells directly via melanopsin or channelrhodopsin channels, or indirectly via opening of neighboring LTCCs on the plasmalemmal membrane. Work is currently on-going to determine if MEMRI is also useful to study cone photoreceptors using zebrafish and coneonly mice (Klaassen et al., 2011; Samardzija et al., 2014).

6. Non-MEMRI assays of rod cell function

Disease or aging will alter a variety of rod functions and not just rod LTCCs. Thus, two additional MRI indices of rod cell function—measured without administration of exogenous contrast agents — have been developed to complement MEMRI and to facilitate translation of MRI of retinal function into humans. These methods are i) apparent diffusion coefficient (ADC) MRI, which measures light-evoked expansion of both SRS [controlled in part by the photoreceptors and in part by the retinal pigment epithelium (RPE)] and of the choroid (Berkowitz et al., 2015a; Bissig and Berkowitz, 2012), and ii) measurement of the spin-lattice relaxation rate in the rotating-frame (1/T1ρ) which evaluates light-stimulated rod arrestin and its translocation, together with retinal vessel expansion (Berkowitz et al., 2015c). As summarized in Fig. 4, in diabetes, for example, abnormal rod function was measured on MEMRI, as well as on ADC MRI and in the light-evoked responses of the retinal vessels and choroid; anti-oxidant treatment corrects most of these defects (except choroidal expansion), supporting a role of oxidative stress in rod cell dysfunction (Table 1). Details about these additional methods are presented below. These results support triangulating the rod cell dysfunctionome using a variety of complementary functional MRI indices.

Table 1.

Anti-oxidant approaches so far examined that correct early diabetes-induced MRI retinal abnormalities.

• α-lipoic acid corrects  • Impaired autoregulation (IOVS, 2006 Sep; 47(9):4077-82)  • Subnormal rod manganese uptake (IOVS, 2007 Oct; 48(10):4753-8)  • Impaired subretinal space expansion (IOVS, 2015 Jan 8; 56(1):606-15) • Cu/Zn SOD overexpression corrects  • Subnormal rod manganese uptake (IOVS, 2009; 50(5):2351-8) • Peroxisome-targeted catalase corrects  • Subnormal rod manganese uptake (IOVS, 2015; 56(5):3095-102) • Retinaldehydes correct  • Subnormal rod manganese uptake (Mol Vis. 2012; 18:372-6)  • Impaired subretinal space expansion (in press, 2015)

6.1. ADC MRI

ADC MRI provides an index of water self-diffusion in a particular diffusion-weighted magnetic field gradient direction. ADC can be thought of as a barrier index: the more barriers water hits as it diffuses, the lower the water mobility, and thus the lower the ADC measured. In the retina, advantage is taken of this property and the fact that the subretinal space (SRS) expands several fold in light compared to that in dark (Adijanto et al., 2009; Chen et al., 2008, 2011; Li et al., 1994a; Nair et al., 2010). In-line with this physiology, ADC MRI measures a greater axial ADC in the SRS in the light than in dark (Bissig and Berkowitz, 2012); the same extent of light-stimulated SRS expansion is measured independently of the presence of manganese (Fig. 11). The sensitivity of ADC MRI to rod function was confirmed by the lack of an increase in the SRS ADC in the absence of rod phototransduction in GNAT1^−/− mice (Fig. 11) (Bissig and Berkowitz, 2012). Here, we present new data, based on literature studies of SRS hydration physiology (Wolfensberger et al., 1999), that predict that the light-evoked increase in ADC in the far posterior retina would be absent if the SRS was acidotic. Pretreatment with a carbonic anhydrase inhibitor acetazolamide, which targets the retinal pigment epithelium (RPE) and induces retinal acidosis, prevented the light-evoked increase in SRS ADC (Fig. 11) (Yamamoto and Steinberg, 1992). These studies support interpretation of ADC MRI of light-dependent SRS expansion as a measure of a combination of rod and RPE function.

Fig. 11.

Light detection using ADC MRI. A) Summary of central retinal ADC with retinal depth during dark (closed symbols, n = 23) and light (open symbols, n = 23) in untreated mice (WT). Approximate location of retinal layers is indicated (dotted blue lines and insert cartoon). Profiles are spatially normalized to retinal thickness (0% = vitreous/retina border, 100% = vitreous/choroid border). Red bracket/horizontal line, P < 0.05. B) Summary of paired data (filled = dark, open = light) of WT (n = 23), MnCl₂-treated mice (WT + Mn, n = 6), GNAT1−/− mice (n = 6), untreated + vehicle treated diabetic mice (STZ, n = 9), diabetic mice treated acutely with the anti-oxidant a-lipoic acid (STZ + LPA, n = 8), on-diabetic acetazolamide-treated mice (AZM, n = 8) (Berkowitz et al., 2015a).

The contribution of oxidative stress to SRS hydration physiology has not been previously investigated. As mentioned above, diabetes produced an impairment in the light-evoked SRS/ADC increases (Fig. 4) that was corrected by treatment with a potent antioxidant, α-lipoic acid (LPA) (Table 1) (Berkowitz et al., 2007c). These data support ADC MRI for monitoring the impact of an oxidative stress environment on rod/RPE function.

ADC MRI also provides information about the thickness of the major blood supply to the rod cells, the choroid. This is important because techniques that simultaneously image co-localized photoreceptor structure/function and choroidal structure/function in vivo have not been available. Most previous MRI studies measure choroidal thickness in vivo using an intravascular contrast agent (Berkowitz et al., 2007a; Shih et al., 2013). Instead, ADC MRI nulls the choroidal blood signal (essentially very fast moving water) in the axial direction allowing a measurement of a “retina”-only thickness (Bissig and Berkowitz, 2012; Chen et al., 2008). Subtracting the “retina”-only thickness from the “retina/choroid” thickness, calculated from non-diffusion weighted images, yields the choroidal thickness. Agreement was found between diffusion weighted choroid thickness and those in the literature thus supporting the accuracy of this approach in C57BL/6J mice (Fig. 4) (Barber et al., 2005; Muir and Duong, 2011). As an additional accuracy check, light-adapted Sprague–Dawley rats agreement was noted between ADC MRI measures of choroidal thickness (63 ± 8 μm, n = 5, mean ± SEM) and a previous estimate (66 ± 8 μm) obtained using an exogenous vascular contrast agent (Berkowitz et al., 2007d).

The choroid has an extensive autonomic innervation that controls its volume in response to changes in light, among other factors (Fitzgerald et al., 1992, 1996; Fuchsjäger-Mayrl et al., 2001; Fuchsjäger-Mayrl et al., 2003; Garhofer et al., 2002; Longo et al., 2000). Indeed, significant light-evoked expansion of choroidal thickness in mice was observed using ADC MRI (Berkowitz et al., 2015a). To assess sensitivity, it was noted that light-adapted choroidal thickness and its regulation are reduced with diabetes and age as has been reported in other studies using other methods (Dallinger et al., 1998; Fitzgerald et al., 2001; Ko et al., 2013). This also occurs in mice; the diabetes results are presented in Fig. 4. In the light, the choroidal thickness in 2 mo C57Bl/6J mice (86 ± 3 μm, n = 18) was significantly (P < 0.05) greater than that in 10 mo C57Bl/6J mice (66 ± 5 μm, n = 4); dark values did not change with age (62 ± 2 μm vs. 59 ± 7 μm, respectively). ADC MRI thus provides assays of photoreceptor/RPE function together with choroidal thickness and its light-evoked expansion that can be combined with MEMRI.

6.2. 1/T1ρ MRI

1/T1ρ, the spin-lattice relaxation rate constant in the rotating frame, describes the decay of the transverse magnetization out of a spin-lock radio-frequency field. 1/T1ρ is particularly sensitive to changes in (i) macromolecular content in avascular tissues as well as to (ii) vascular volume in well perfused tissue like the brain (Hulvershorn et al., 2005; Jin and Kim, 2013; Kettunen et al., 2002; Li et al., 2007; Magnotta et al., 2012; Regatte et al., 2003). 1/T1ρ has been used clinically (Borthakur et al., 2004; Haris et al., 2015; Hulvershorn et al., 2005). Because of its sensitivity to large molecules, it was hypothesized that the difference between avascular outer nuclear/inner segment tetrameric rod Arrestin 1 (Arr1) in the dark – and its reduction via translocation upon light exposure – would be detectable as a change in 1/T1ρ in the outer retina (Gurevich et al., 2011; Huang et al., 2010). Tetrameric Arr1 plays an important role in regulating rod synapse function (Gurevich et al., 2011; Huang et al., 2010). Exposure to bright light produces a dramatic reduction in tetrameric Arr1 concentration in these compartments with translocation of the monomer form of Arr1 to the outer segments; only Arr1 monomer binds rhodopsin to stop phototransduction (Hanson et al., 2007; Philp et al., 1987). Importantly, this translocation of Arr1 into the outer segments does not occur unless visual cycle regeneration of rhodopsin via the visual cycle is normal (Mendez et al., 2003). As shown in Fig. 4, 1/T1ρ transretinal profiles clearly demonstrate a greater 1/T1ρ in the dark than in the light (P < 0.05). The opposite pattern was noted in the inner retina (i.e., <50% depth into the retina) (P < 0.05); these results were independent of the presence of manganese (Berkowitz et al., 2015c). Confirmation that this change in 1/T1ρ in the rod cells was provided by the fact that the arrestin 1 signal at 64–72% was absent in Arr1^−/− null mice (kind gift from Dr. Jeannie Chen, USC) and greater than normal with Arr1 overexpression (data not shown) (Berkowitz et al., 2015c).

1/T1ρ in blood is known to be much shorter than in tissue (Hulvershorn et al., 2005; Jin and Kim, 2013; Kettunen et al., 2002; Magnotta et al., 2012). This allowed an evaluation of light-dependent changes in the inner retina vascular volume based on changes in 1/T1ρ (Fig. 4) (Hulvershorn et al., 2005; Jin and Kim, 2013; Kettunen et al., 2002; Magnotta et al., 2012); the light-evoked change in inner retinal blood volume was confirmed using MRI and the blood-pool contrast agent monocrystalline iron oxide nanocolloid (MION) (Shih et al., 2011). Supporting this finding, our lab finds no photoregulation of 1/T1ρ in the inner retina in (i) guinea pigs (which are without an inner retinal circulation, data not shown) and (ii) diabetic mice (Fig. 4). The data in the diabetic mice are particularly relevant because diabetes is known to impair retinal vascular autoregulation (Kern et al., 2010). Since the retinal circulation provides ~10% of the oxygen required by the photoreceptors in the dark to support the great energy needed for continuous ion channel opening (41–44), evaluating the photoregulatory changes in inner retina blood volume is likely useful as an indirect probe of functional changes in photoreceptor oxygen consumption.

Using a combination of MRI approaches, rod compartment-specific LTCC subtypes can be studied together with light-stimulated changes in SRS, tetrameric Arr1, choroidal thickness, and retinal vessel autoregulation in vivo. To ensure spatial accuracy in vivo, high resolution MRI assays of rod function can be aligned against optical coherence tomography (OCT) images.

7. MRI and oxidative stress

Oxidative stress is well known to impair neuronal function, including LTCC function, suggesting the hypothesis that MRI indices are sensitive to the presence of oxidative stress (Downs and Helms, 2013; Fusi et al., 2001; Guerra et al., 1996; Guzman et al., 2010; Kourie, 1998; Shirotani et al., 2001; Yang et al., 2013). All of the published MRI data to-date are consistent with this hypothesis. For example, in 2007, it was observed that ouabain, a specific inhibitor of Na⁺/K⁺-ATPase activity that increases neuronal oxidative stress, significantly reduced uptake of manganese in rod cells as measured by MEMRI (Riegel et al., 2010). This lower-than-normal pathophenotype was also found early in many diseases with a prominent oxidative stress etiology, including models of DR, RP, glaucoma, and ischemia/reperfusion (Campochiaro et al., 2015; Du et al., 2013; Fukuda et al., 2014; Handa, 2012; Inman et al., 2013; Osborne et al., 2004; Roddy et al., 2012); the ADC MRI and 1/T1ρ MRI-measurement of autoregulation to retinal oxidative stress data were presented above (Berkowitz et al., 2007b, 2011, 2013, 2008b; Calkins et al., 2008). Intriguingly, in a rat model of retinopathy of prematurity, a model not associated with oxidative stress, a greater than normal uptake in rod cells and cells of the inner retina was observed; these results are not consistent with oxidative stress but instead suggest a loss of HC inhibitory signaling (Berkowitz et al., 2007b, 2011). To be clear, simply observing multiple aspects of dysfunction linked with oxidative stress is not sufficient to use functional MRI indices as a tool for the unambiguous detection of oxidative stress per se since subnormal uptake may arise for reasons unrelated to oxidative stress. However, and most importantly, a range of treatments known to reduce retinal oxidative stress also correct the MEMRI defects in DR (Table 1); in DR, antioxidant treatment also corrected the early defect measured by ADC MRI (Berkowitz et al., 2015a).

Together, these data suggest that MRI is a powerful approach for objectively evaluating antioxidant treatment efficacy in a rod cell compartment-specific manner before the appearance of other clinical biomarkers in disease in vivo.

8. New perspectives

Aging:

Although many retinal diseases occur against a background of aging, the influence of the aging process on disease progress remains vastly underexplored. Aging alone causes declines in vision even in the absence of disease and available biomarkers have been insufficient in explaining the physiology behind these declines (Fitzgerald et al., 2001; Gresh et al., 2003; Grunwald et al., 1993, 1998; Kolesnikov et al., 2010; Lehmann et al., 2012; Spear, 1993). While the extent of vision loss with age in healthy patients is modest, its occurrence alone is a significant predictor of degrading day-to-day quality of life, deleterious aging outcomes (e.g., hip fracture), and reduced survival (Swindell et al., 2010). Thus, an increasingly older population raises an urgent and continuing need to understand the origins of age-related physiologic vision loss and how this impacts disease-related vision loss (Bissig et al., 2013). Until recently, biomarkers of retinal physiology have been insufficient for understanding how the “healthy” retina ages (Spear, 1993). In particular, the role of retinal LTCCs and their subtypes in age-related vision loss has been virtually unexplored. A strong predictive link between age-related escalating increases in photoreceptor Ca_v1.3 LTCC activity as measured by MEMRI and progressive declines in contrast sensitivity has been observed (Bissig et al., 2013). In support of this hypothesis, healthy C57Bl/6J mice have relatively stable photopic vision until after 18 mo of age (data not shown) (Bissig et al., 2013; Kolesnikov et al., 2010; Lehmann et al., 2012; Spear, 1993) and non-escalating retinal LTCC function, as measured by MEMRI, with light and dark adaptation before 18 months of age. Together, these data have uncovered a previously unsuspected target in age-related impaired sight loss: early and progressive increases in retina LTCC function; intriguingly, this calcium hypothesis of aging is reminiscent of a similar hypothesis first identified in CA1 region of hippocampus and now linked with cognitive declines (Bissig and Berkowitz, 2014). As noted elsewhere, these aging changes may help explain some of the later time course features in DR as measured by MEMRI (Berkowitz et al., 2013).

9. Addressing concerns about MEMRI and MRI

A considerable body of work over the past decade has demonstrated that MRI has the spatial resolution and sensitivity to be a reliable and physiologically accurate technology for analytically measuring rod cell biology non-invasively in a range of animal models. Here, we address some of the major concerns.

MEMRI requires injection of manganese, an essential metal at low levels (e.g., manganese-superoxide dismutase) but a neurotoxin at high levels of exposure. For MEMRI to be useful it is critical to identify doses that are non-toxic and that allow for adequate detection of manganese with good contrast-to-noise on MRI examination. At the doses used in studies of rodent retina [44 (rat) – 66 (mouse) mg/kg, IP], no evidence was observed that manganese altered retinal health based on lack of changes in 1) light adaptation, 2) synaptic release at the rod synapse, light-evoked expansion of the 4) subretinal space, 5) choroid, and 6) retinal vessels, 7) arrestin 1 and its light-dependent translocation, 8) blood retinal barrier integrity, 9) intraocular pressure, 10) electroretinogram parameters (ERG) and 11) histology (Berkowitz et al., 2006, 2007a, 2007b, 2015a, 2015b, 2015c). Here, new data are presented that neither 12) visual acuity nor 13) contrast sensitivity were altered 4 h post injection of manganese (66 mg/kg MnCl₂ treatment in dark adapted C57Bl6 mice IP), as measured by optokinetic tracking (P > 0.05) (non-treated (n = 5) acuity 0.389 ± 0.004 cycles/degree and contrast sensitivity 16.9 ± 1.2 unitless parameter; treated (n = 5) acuity 0.394 ± 0.005 c/d and contrast sensitivity 19.2 ± 1.7 unitless parameter); also, no changes from baseline values were observed at 24 h and 1 week post MnCl2 treatment (data not shown). In addition, we evaluated 14) rod-dominated retinal superoxide production with a lucigenin assay and confirmed similar (P > 0.05) levels in untreated mice (n = 3,100.0 ± 10.9% of controls after normalizing for protein (mean ± SEM)) and mice 4 h post treatment with manganese (66 mg/kg, IP; n = 3,102.6 ± 18.7% of controls after normalizing for protein). Collectively, these 14 different indices strongly suggest a lack of toxic effects of manganese at the singly acute systemic doses of MnCl2 used in retinal studies.

A related concern is whether or not MEMRI can be safely applied in humans. To this end, note that the Mn²⁺ injectate TeslaScan is a manganese-based contrast agent previously approved over a decade ago as safe and effective for use in humans by the Food and Drug Administration (FDA) (Karlsson et al., 2015; Tofts et al., 2010; Torres et al., 1997). Teslascan was originally investigated as a blood pool contrast agent to identify the tumor burden in the liver. Teslascan has shown to lack adverse events, even in patients with cancer and patients that needed cardiac reperfusion (Karlsson et al., 2015). Teslascan is converted in vivo into the lipid soluble Mn dipyridoxyl ethyldiamine (Karlsson et al., 2015). In addition to not showing toxicity, Teslascan and its metabolite act as anti-oxidants because they are superoxide dismutase mimetics, and have ironbinding abilities in vitro (Karlsson et al., 2015). Preclinical and early clinical studies confirm their beneficial protection against oxidative stress in different tissues (Karlsson et al., 2015). To determine if the low concentration of chelated manganese in Teslascan would be useful in functional studies, it was observed that the Mn²⁺ in Teslascan is taken up by the rat retina in a stimulus-dependent manner (Berkowitz et al., 2013; Tofts et al., 2010). Together, these considerations have motivated on-going human MEMRI studies using Teslascan to detect accumulation of Mn²⁺ with activation in retinal and brain neurons. Note, however, that the Teslascan used in this clinical study was prepared in-house because there are currently no commercial sources of Teslascan due to a low market earning (Karlsson et al., 2015). Thus, while feasible and safe in patients, wide application of MEMRI likely awaits improved availability of Teslascan.

Can the high resolution functional images similar to those described in animals models herein be obtained in patients? Spatially the problem is simply defined: given a 250 μm thick retina, one needs 50 μm or less resolution to collect enough information to confidently separate inner from outer retina. By taking advantage of the good coil filling factor and non-isotropic sampling (e.g., thick slices), fine resolution images, approaching those in rodents, but in humans without sedation is possible using a cued-blinking procedure (Beenakker et al., 2013, 2015; Berkowitz et al., 2001; Berkowitz et al., 2013; Lindner et al., 2014; Sirin et al., 2015; Zhang et al., 2011). For example, we recently published human images collected at 58 μm axial resolution (Berkowitz et al., 2013). The possibility of using MEMRI in patients was discussed above. ADC MRI requires no exogenous contrast agent and is likely to be useful in human studies. Clinical application of 1/T1ρ MRI of brain tissue in patients is already in place and FDA limitations for specific applied radiation (SAR) is possible with available acquisition schemes (Borthakur et al., 2004; Haris et al., 2015). Thus, while challenges undoubtedly remain in making high resolution functional retina images in patients routine, the results to-date clearly highlight likely progress in this area.

Will MRI be useful as a screening tool in patients? Not likely because MRI throughput is relatively low, and MRI facilities are located outside of vision labs and Ophthalmology clinics. On the other hand, these are not limitations if MRI is used as a discovery tool in, for example, animal studies, Phase 2 clinical trials or in patients with difficult to manage disease. Most major hospitals have clinically accessible MRI centers, and the recent availability of relatively low cost bench top preclinical MRI systems raises the possibility that more labs will be able to routinely benefit from MRI in their laboratories.

The expense of MRI is also considered to be problematic. However, it is important to consider not just the cost but the cost-benefit ratio. For example, the equipment used for non-MRI methods is considerably less expensive than a MRI machine; most major research centers, however, have small animal MRI available as core facilities. The benefit of existing non-MRI methods, as discussed above, is less clear because they provide incomplete and incompatible assays that do not measure many essential aspects of retinal function; thus, current methods have a low cost but also a low benefit. To see why the low cost/low benefit approach is problematic, we note that it is standard practice to wait until the appearance of gross histopathology to implement laser therapy; this is clearly costly to the retina and patient. Applying existing tools to expensive genetically modified mouse models and genetic therapies is also very costly because these popular approaches have not shown good predictive value for future clinical success (e.g., (McGill et al., 2012)). On the other hand, MRI provides a unique discovery engine that measures a wealth of retinal and rod compartment-specific functions in the same eye at the same location at the same time, that are unavailable with other technologies, resulting in much higher benefits for directed, personalized, and targeted therapy.

10. Summary and future directions

It is currently a major challenge to translate findings in animal models into the clinic in part because of vast differences in methodology and available indices. The best bench-to-bedside (and back) bridge is likely to be the development and application of imaging biomarkers that inform about disease progression and treatment efficacy in both preclinical and clinical situations. It has become clear that current optical and electrophysiology methods do not provide such a bridge in many situations. This problem can perhaps be best appreciated early in the course of retinopathies (i.e., in the absence of gross anatomy changes), a critical time point when drug intervention is most likely to be successful at reducing vision loss. In other words, before the appearance of gross changes there is currently little to guide the management of retinal disease and to answer essential questions such as what is the optimal time/dose/route/schedule for therapeutic intervention? Is a treatment reaching its target? How well do current experimental models mimic the clinical condition?

This review highlights MRI as a promising tool for addressing the above problems. High resolution MRI stands alone in its ability to analytically measure rod cell compartment-specific function in vivo. MRI is thus a modern phenotype discovery tool that can complement genotyping assays. The availability of this optimized imaging technology enables earlier evaluation of disease progression and better management of disease-preventing anti-oxidant treatment than is currently possible. Most of the newer assays of rod cell functions are based on endogenous contrast mechanisms, and this is expected to facilitate their translation into patients with DR and RP, and other oxidative-stress-based retinal diseases.

MRI studies uniquely allow for in vivo testing of ex vivo-derived hypothesis and this has already motivated new lines of scientific inquiry into, for example, the problem of vision loss in aging and in DR. In addition, MRI provides new and useful biomarker of therapeutic efficacy against threats to sight. As MRI studies of rod photoreceptors become more common used by other labs, better testing of mechanistic hypotheses and early evaluation of treatment efficacy in vivo is expected.

Abbreviations:

ADC

apparent diffusion coefficient; Arr1, Arrestin 1

ASL

arterial spin labeling

BRB

blood retinal barrier

BOLD

blood-oxygen level dependent

Cx36

connexin 36

Cx57

connexin 57

DIL

D-cis-diltiazem

DR

diabetic retinopathy

ERG

electroretinogram

HC

horizontal cell

IS

inner segments

MEMRI

manganese-enhanced MRI

OCT

optical coherence tomography

ONL

outer nuclear layer

RPE

retinal pigment epithelium

RP

retinitis pigmentosa

1/T1r

spin-lattice relaxation rate in the rotating-frame

SRS

subretinal space

LTCCs

voltage-gated L-type calcium channels

wt

wildtype.