There are a growing number of hereditary and degenerative diseases of the retina that are the target of pre-clinical or clinical research using both genetic vectors or cell based therapies (1, 2).

The use of viral vectors in humans has centered on understanding the mechanisms of disease leading to a progressive loss of photoreceptors through apoptotic and non-apoptotic mechanisms, as well as the means to correct the defect (3, 4).

These treatments are dose and volume dependent, somewhat similar to pharmacologic therapy. Hence, optimizing delivery, devising a reproducible, standardized method to appropriately target retinal tissues and layers can improve on the experimental and clinical results by reducing the loss of vectors from the intended target and a reduce risk of an immune response.

It has been assumed that current available surgical techniques are sufficient and reliable in the hands of surgeons, many of whom, have limited training in performing this type of surgery (5, 6). Several lines of evidence indicate that this may not be the case (7–9).

Three surgical approaches are currently proposed to reach the retina: direct intravitreal injection which fills the core of the eye and exposes all intraocular surfaces in contact with the vitreous to the vectors or cells, or by injecting the suspension under the sensory retina either by a transvitreal route or reaching the subretinal space via the suprachoroidal space, a virtual space between the sclera and deeper eye structures (Figure 1) (10, 11). Each approach brings a unique set of advantages and drawbacks which should be considered in the context of the chosen therapeutic strategy and adapted to each specific disease being targeted (Table 1).

Accessing the eye by intravitreal injections or via the suprachoroidal space was proposed to circumvent the difficulties inherent with a transvitreal approach for subretinal delivery.

This review will address the challenges, advantages, and limitations of the current methods used to target the subretinal space. We will explore the role of robotic technology in understanding, optimizing and delivering gene and cell products to the subretinal space and propose solutions that make use of this novel technology.

Altered, deviant immunity present in the subretinal space provides a certain degree of protection from inflammation and rejection but it does not prevent exposure to immunosurveillance mechanisms. Allogeneic fetal or pluripotent stem cells have both been known to cause an immunogenic response when injected intraocularly (37). The immune response varies depending on the mode of delivery (27). Indeed, clinical trials conducted using allogenic RPE cells for the treatment of AMD, 2 decades ago, were all met with graft rejection (38–42). Immune responses are also commonly observed using viral vectors and with mRNA transfections (42–44). The clinical features of these rejection episodes are often subtle and depend on the mode of delivery. They include inflammation in the anterior chamber and the vitreous, depigmentation at the site of injection, serous retinal detachment or retinal edema, mild to moderate cell infiltration around the cell transplant or in the overlying vitreous (seen on OCT) as well as mild vasculitis on fluorescein angiography (Table 2). Following an intravitreal injection, a dose dependent immune response is observed presenting as cells in either the anterior chamber and/or the vitreous. Both cellular and humoral factors have been implicated (45–47). In mice, an AAV injection into the vitreous leads to a transient mild spontaneously resolving inflammation but the total number of CD45+ T cells remains elevated, even weeks after the injection. Both innate and adaptive immunity play a role regardless of prior immune status (45). A patient treated for Lebers with a single intravitreal dose of 9 × 10¹⁰ vg and high pre-treatment titers of IgG neutralizing antibodies to AAV2, developed a significant post-injection ocular inflammation (46). Similarly, in April 2021, a patient enrolled in the INFINITY trial using Adverum's ADVM-022, an AAV.7m8 vector encoding a sequence for aflibercept developed hypotony, panuveitis and vision loss, 30 weeks following injection of 6 × 10¹¹ vg (48). While improved pre-injection screening may reduce the incidence of such events, they limit the applicability of this approach. The presence of inflammation also will limit the efficiency of transfection or the duration of the effect if it leads to the elimination of transfected cells (45, 49, 50).

In contrast to the intravitreal route, transvitreal subretinal injections elicits less of an inflammatory reaction (51–53). Dose dependency has been observed both experimentally and clinically (54). In the Voretigene trials, 8% of patients showed signs of transient inflammation (52). At higher tiers and volumes, mild vitritis, optic disc swelling and some sheathing was observed several weeks after administration, and some focal pigmentary changes were observed 3.5 years later in the same patient (54). Hyper reflective foci were observed transiently in the retina of a patient treated with a low dose of AAV8, while at the intermediate dose, transient iridocyclitis was seen (55). Experimental studies confirm the lower immunogenicity of the transvitreal subretinal route but confirm that at higher doses (1 × 10¹² vg for AAV8), inflammatory cell infiltration in the retina and choroid are observed in non-human primates (56). Subretinal injection do not seem to induce antibody production (57), also confirmed in another study by the group of R Ali, but only for low dose AAV injections. Indeed, higher doses lead to the expression of neutralizing antibodies that reduce the efficacy of repeated vector administration (58). At higher doses, the risk of a vitreous reflux increases, and a significant number of viral particles can persist despite flushing of the vitreous cavity with saline or BSS (59). Subretinal injections using transcleral micro needles lead to a diffuse peripheral expression when injected close to the pars plana but was associated with a localized inflammatory response consisting in the accumulation of macrophages and causing a localized chorioretinitis (60). When present, these responses appear 1–3 months after the therapy was applied (11). In the presence of inflammation, gene expression was noted to decrease progressively. Of note, even expression of AAV8 in the suprachoroidal space can be associated with a mild chorioretinitis and altered photoreceptor morphology (61).

Subretinal and intravitreal injections appear to lead to different phenotypes of immune response. With current vectors, the safest approach from an immune standpoint appears to be subretinal delivery, though this approach may not be appropriate if more superficial retinal cells such as ganglion cells are the targeted cell type. While the risk for an overall population may be small from an intravitreal approach, an immune response if induced can have devastating effects on vision as recently shown. These risks can be minimized by direct intraretinal targeting which can avoid priming the immune system and make use of the deviant immune response inherent to the retina and sub retinal space.

Recent advances in imaging and robotics can overcome the limitations listed above (7, 22). Robotic systems have been in routine use for more than 20 years in other surgical specialties. Advances in microrobotics make it now possible to perform highly delicate and precise surgeries such as the anastomosis of lymphatic vessels (62, 63). However, in gene / cell therapy, the inclusion of precision robotics to deliver accurately to a target location has only been explored to a limited extent. In obstetrics an article of 2016 has reported its successful use of a robot for the transplantation of frozen-banked ovarian tissue, and a proposal in 2021 suggests the use of a robotic platform to deliver stem cells into the brain of patients with Huntington's disease. The potential use of robots to treat other degenerative diseases is stimulating research into novel robotic platforms with for example MRI guidance of catheters to the spine, or even in cardiac surgery (64–67).

In ophthalmology, high precision and accuracy are also required in the delivery of cell or gene products, and robotics can offer a solution. The terms precision and accuracy are often misunderstood. Precision refers to the degree of reproducibility of a motion, while accuracy refers to the contiguousness achieved in reference to an intended target This is best exemplified in Figure 3. Accuracy and precision can also be defined as components of both dynamic and static tasks of which the latter are more difficult to maintain. They can also be defined as a function of the axis (XYZ), and as shown in simulation experiments, the Z axis is the most demanding and where experience makes a difference (14). Unaided, dynamic precision in XY is about 40 μm for experienced and novice surgeons, and 35 vs. 60 μm in Z. Accuracy under the same conditions is between 68 and 87 μm for XY and 58–108 μm in Z. Robotics when self-actuated has an accuracy and precision between 1 and 3 μm.

Another issue is the location of the center of motion (RCM). Ideally it is placed at the site of insertion into the eye, but existing systems often have their RCM at a separate location (68). In Intuitive's da Vinci system, the RCM is located away from the eye making all intraocular movements less controllable and promotes unnecessary tension at the surface of the eye (69). Adapting existing robotic systems to ophthalmology is therefore fraught with problems (70).

Robotic systems specifically designed for intraocular surgery fall in three main categories: smart surgical tools, comanipulation and telemanipulation which are described more extensively in a review from 2018 (71). Table 3 summarizes advantages and drawbacks of different robotic designs for eye surgery in regard of subretinal delivery requirements. Master-slave systems allow a decoupling between the manipulation of an instrument from the surgeon's direct grip. Particularly in tele-manipulated systems, where the movement of the slave is controlled by a computer, this enables additional functions such as tremor filtering and an ability to introduce a variety of other commands that can lead to the precise positioning of the tip of a catheter at the appropriate depth under the retina. A catheter placed at the retinal surface, after an appropriate assessment of retinal thickness on an intraoperative OCT (iOCT), can be advanced to the exact required distance to place the tip of the instrument at the retinal RPE interface. It is the standby functionality the ability to suspend any task carried out by a robotic system that allows such fine measurements and provides the surgeon with the ability to carry out these precision task, uninhibited by time constraints. This independence means that the advancement of the needle through the retina can be planned once inside the eye in real time using existing iOCT machines, and because there is little motion of the eye, the signal to noise compensation algorithms are fully functional which allow for a fully optimized image of various retinal planes (72).

Robotics also gives researchers the ability to fully dissect a surgical maneuver to determine for subretinal injections for example the appropriate angle of penetration, depth, retinal contour and speed of injection that minimizes or prevents the risk of reflux. Such understanding and optimization were not possible prior to the use of robotics. By automating some of the steps, limiting the surgeon's interaction to steps which require his/her expertise such as choosing the appropriate location in the posterior pole and an adequate positioning at the retinal surface, the critical steps for the delivery of the cells or gene product to the subretinal space can be standardized (Figure 4).

This ability to reduce variability was clearly shown in an experiment carried out with retinal surgeons of various skill levels at a European retinal meeting (Euretina). Vitreoretinal surgeons, who had never used the PSS were asked to perform a simulated subretinal injection with and without the robotic device (7). A bleb was created more frequently with the use of the PSS (88 vs. 44%) with a reduction in the rate and severity of reflux (77 vs. 88%) was observed in this model that lacked any elastic tissue. Tremor was clearly reduced (Supplementary Video 1) when using the robot. The ability to hold the instrument steadily at the point of insertion ranged from 40 to 266 μm, depending on the individual surgeon, when surgeries were performed manually as compared to, 1–2 μm with robotic assistance (Figure 5) resulting in a diameter size reduction of the retinal hole. The use of the robotic arm also led to a longer “infusion time” ranging from 13 to 108 s, while the injections performed manually ranged from 18 to 85 s. In both cases, they had been instructed to inject over 20 s. These observations confirm studies performed in ex-vivo porcine eyes and in vivo data (73). In this same ex-vivo model, the use of PSS lead to the formation of subretinal blebs in 100% of cases, and reflux limited to 20% while manually, a bleb could only be created in 40% of cases with 100% reflux as seen on iOCT. Of note, to enhance visualization of reflux a contrast agent was added to the injection solution which allowed for the clear identification of reflux (Figure 6) (25).

Microprecision robotic systems will require the key characteristics if they are to be used for ocular gene/cell therapy:

While many of these points have been demonstrated in a non “gene therapy” setting (74), it is necessary to demonstrate the value in gene/cell therapy applications. Optimization of several functions will require further optimization with regards to retinal location and thickness. As each application and delivery location will require its own set of parameters. In essence for cell type, disease entity, intraocular location, specific proprietary software and hardware (intelligent devices) can be created.

The initial step will be to optimize parameters in live animals (pigs, rabbits, or monkeys) so that each injection achieves 90–95% of the objectives 1–5 listed above.

In a follow-up stage, incorporation of distant sensing and later pressure sensing would allow automation of the procedure. If coupled to an automatic infusion line, it could dynamically adapt to the degree of retinal stretch or relaxation as the bleb develops. In healthy retina, this is of course not needed, but ultimately, not only healthy retina of patients with genetic diseases will be targeted but also patients with thinned or scarred retinas, where it is difficult to appreciate the degree of tissue plasticity. At that stage, these further refinement in delivery may show that a full vitrectomy is superfluous.

Safe, efficient and reproductible subretinal delivery of gene vector/cells solutions require skills which are literally at the limit of human dexterity.

Robotic assistance especially highly versatile telemanipulation robots can overcome these barriers. Such a tool would standardize the surgical procedure, increasing accuracy and precision resulting in a higher efficiency and safety and therefore better outcomes. It would also reduce the cost of clinical trials as the variability of drug delivery between surgeons as well as between centers will be significantly reduced. Costs of pre-clinical in vivo experiments can also be diminished for the same reasons. A standardized delivery system also facilitates the adoption of an approved drug as training of surgeons in the use of this delivery device can be limited in time and space and can be carried out with phantoms rather than live animals or ocular tissue.

All authors listed have made a substantial, direct, and intellectual contribution to the work and approved it for publication.

FW has been part of advisory boards for Abbvie, Allergan and Santen. MdS is Chief Medical Officer of Preceyes B.V., Eindhoven, Netherlands. He has received honoraria from Allergan, has been a part of advisory boards for Abbvie, Allergan, Janssen and Oxular Ltd. MdS was employed by MIOS SA. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.