RetINaBox: A Hands-On Learning Tool for Experimental Neuroscience

Synthesis

There is a agreement among the reviewers that the work is potentially impactful and that the tools presented here can be useful for engaging students in learning about the retina. There are also several important concerns regarding the oversimplification of some aspects of retina function in the tool design. While some simplification of function may be necessary for teaching, the reviewers agree that in the current form the reduced complexity may promote incorrect conceptual understanding of retina function. Specific comments are provided below.

Major concerns

The creators of "RetINaBox" have tackled a very important area in which there is a great need for pedagogical innovation. As they beautifully put it in their Introduction, "A challenge in neuroscience education is how to incorporate such experimental work into lesson plans, so that instead of solely learning lists of facts, students get hands-on experience that captures the excitement of discovery." The introduction raises the appropriate excitement about the tool and the goals of the study. However, the current form of device does not meet the expectation.

When reading the details of the tool the realization was that it may be excessively simplified: 9 photoreceptors hard-wired (essentially) to 2 retinal ganglion cells seems insufficient to provide insight into fundamental aspects of retinal function. While it is clear that the creators have put much hard work and love into their creation and some simplification is necessary and, often, pedagogically useful, there is the concern that students would come away from the lessons with a distorted view of how the retina actually works. Just to take one example, the elegant circuitry of the inner nuclear layer enables every photoreceptor to contribute to both 'on' and 'off' channels (using the ingenious device of sign-conserving vs. sign-inverting glutamate receptors in bipolar cells) and to participate in the 'center' and 'surround' of different retinal ganglion cells. Moreover, because the horizontal cells provide inhibitory feedback to the photoreceptors, the circuitry allows the same neurons to provide opponent spatial surrounds for both 'on' and 'off' channels. The concern here is that students would come away thinking that each photoreceptor has a direct connection to an RGC and that it has a dedicated role-i.e. either center or surround. Much of the beauty of the retina is in the details of the circuitry.

What educational level this tool would be appropriate for? In their closing sentence, the authors assert that it would be useful for students in high school to graduate school. In the current configuration it seems it may being useful for younger students, but graduate students would likely find it overly simplistic.

One important component of retina function that typically raises students' engagement comes from freely available movies showing that a large spot of light fails to elicit much response from the 'on' RGC, which is further deepened when the students see the somewhat puzzling suppression by the annulus of light. This provides a great opportunity to then delve into the RF structure and the underlying circuitry. The typical reaction to the movie of the 'end stopped' cell is even better. The point here is that these movies are freely available, and may do a better job of conveying the excitement of discovery than does RetINaBox.

It would be helpful to note whether the design could be scaled (e.g., by using GPIO expanders or multiplexing) to support more complex receptive field demonstrations. Even if the authors do not intend to pursue this now, acknowledging scalability as a possibility could broaden interest and highlight the flexibility of the platform.

Minor issues.

What is the actual estimated cost and the amount of time necessary to make the device. From looking over the user's manual, it seemed that it would be a fairly time-consuming task. It would be helpful if the authors included a general cost estimate and some idea of how long it might take, say, a high school science teacher to build.

There were also a couple of minor things that raises concerns, such as the statement that center-surround RFs of different sizes were for detecting objects of different sizes. Most vision scientists think of center-surround as a mechanism for reducing redundancy and signaling contrast. Different center sizes are important for conveying information about different spatial frequencies, but "object" is a much more complex and nuanced computation. Another was the question setting up lesson 3: "How do we know if and how objects are moving relative to us?" The kind of simple motion detector being built up with pure excitation and phase delays is a far cry from the circuits that detect relative motion.

Author Response

Response to reviewers Synthesis Statement for Author (Required):

There is a agreement among the reviewers that the work is potentially impactful and that the tools presented here can be useful for engaging students in learning about the retina. There are also several important concerns regarding the oversimplification of some aspects of retina function in the tool design. While some simplification of function may be necessary for teaching, the reviewers agree that in the current form the reduced complexity may promote incorrect conceptual understanding of retina function. Specific comments are provided below.

Major concerns The creators of "RetINaBox" have tackled a very important area in which there is a great need for pedagogical innovation. As they beautifully put it in their Introduction, "A challenge in neuroscience education is how to incorporate such experimental work into lesson plans, so that instead of solely learning lists of facts, students get hands-on experience that captures the excitement of discovery." The introduction raises the appropriate excitement about the tool and the goals of the study. However, the current form of device does not meet the expectation.

We thank the editor and reviewers for acknowledging the importance of the topic that our paper addresses. As you will see outlined below in our detailed responses to specific comments, we believe that the revised version of RetINaBox now much better 'meets the expectations' set up in the introduction.

When reading the details of the tool the realization was that it may be excessively simplified:

9 photoreceptors hard-wired (essentially) to 2 retinal ganglion cells seems insufficient to provide insight into fundamental aspects of retinal function. While it is clear that the creators have put much hard work and love into their creation and some simplification is necessary and, often, pedagogically useful, there is the concern that students would come away from the lessons with a distorted view of how the retina actually works. Just to take one example, the elegant circuitry of the inner nuclear layer enables every photoreceptor to contribute to both 'on' and 'off' channels (using the ingenious device of sign-conserving vs. sign-inverting glutamate receptors in bipolar cells) and to participate in the 'center' and 'surround' of different retinal ganglion cells. Moreover, because the horizontal cells provide inhibitory feedback to the photoreceptors, the circuitry allows the same neurons to provide opponent spatial surrounds for both 'on' and 'off' channels. The concern here is that students would come away thinking that each photoreceptor has a direct connection to an RGC and that it has a dedicated role-i.e. either center or surround. Much of the beauty of the retina is in the details of the circuitry.

We thank the reviewers for their concern about possible oversimplification and their concern students could be misled. Below, we respond to these points in detail.

First, while agree with the statement above about the beauty of the retina and about the importance of inner retinal neurons, we do not agree that the simplifications we have made impede learning. Instead, we believe the simplifications we made should aid learning of various concepts in visual neuroscience. To make an analogy: the leaky integrate-and-fire neuron model ignores the beauty of the cell's membrane, ignores the biophysical complexity of ions and ion channels, ignores trying to model the action potential, and ignores many other complexities of actual neurons. Despite all these simplifications, this model has been incredibly useful to the field due to its simplicity.

Second, in the manuscript we now more clearly outline the pedagogical rationale behind RetINaBox, and the specific reasons for the simplifications that were made. We are not trying to build a biologically realistic electronic retina with one-for-one electronic components representing specific retinal cell types. If we were to design a more complex system, it is unclear where we would draw the line between inclusion and exclusion. For example, the reviewers suggest a version where we include ON and OFF bipolar cells and horizontal cells. But then we would need to include amacrine cells for completeness. And it's known that there are multiple types of ON and OFF bipolar cells, amacrine, and ganglion cells, so it's not clear where we would draw the line on how many subtypes of each cell type to include. And the density of many retinal cell types can vary laterally across the retina, so it's unclear how we would incorporate that into a model. Lastly, these myriad cell types form complex cell- type specific circuits and including all of these would be very complicated. Thus, we believe that making RetINaBox much more complex would make it difficult to design and assemble, and make it much less practical as a teaching tool. With RetINaBox, we sought to develop a simple tool that would allow users to explore and learn several important concepts related to computations in the visual system. We decided to focus on ON/OFF, center-surround, orientation selectivity, and direction selectivity, as these are canonical and well-studied, and are often taught in classes and appear in many textbooks. Given that seeing starts in spatially offset light detectors (i.e. photoreceptors), we asked what the minimal set of photoreceptors would be to enable a description of the concepts of ON/OFF, center-surround, orientation selectivity and direction selectivity. This led us to a 3 x 3 array of light detectors. We then developed lessons to let users explore these concepts, via trial-and-error experimentation and discovery, so that we could bring something approximating 'doing experiments in the lab' to the classroom setting. We then ended each lesson with an activity that had users apply the feature selective tuning properties they just learned towards a particular goal (e.g. cracking a code, building a shape detector, etc.). We did so in hope of maximally engaging RetINaBox users to facilitate and reinforce the learning experience. We now more clearly outline this framing of RetINaBox in the manuscript.

Third, to address this reviewer concerns about oversimplification and possible misunderstandings by students, we have overhauled the lesson plans and teaching slides. We now provide specific examples of how the feature selective concepts we explore with RetINaBox (ON/OFF, center-surround, orientation selectivity, direction selectivity) are implemented in real biological circuits, to ensure that users get exposed to the actual circuits in addition to our simplified ones.

Fourth, the reviewers argue that RetINaBox could give the false impression that photoreceptors directly connect to RGCs. We clearly state that this is not the case, and we are transparent about how we have simplified the circuitry. In the updated manuscript and lesson plans we now more frequently emphasize that photoreceptors do not directly connect to RGCs in the real retina.

Fifth, the reviewers argue that RetINaBox could give the false impression that each photoreceptor has a single specific fixed role in the RFs of all downstream neurons. Note that RetINaBox contains 2 RGC output neurons, in large part to make it clear to users that it is the details of the connectivity from photoreceptor to RGC that dictate the RGC's feature selectivity, and that the signal from a single photoreceptor can lead to different outcomes in different downstream neurons. Throughout the lesson plans we frequently ask users to have the 2 RGCs compute similar things at the same time, but in different manners (e.g. in the center-surround lesson we have users make 2 center-surround neurons, but with different center and surround locations; in the orientation selectivity lesson, we have users make 2 orientation selective neurons preferring same line but at different orientations; in the direction selectivity lesson we have users make 2 cells with the opposite preference for a moving stimulus). In each of these cases outlined above, RetINaBox requires that users connect the same 9 photoreceptors to both ganglion cells, but with the connectivity profile (i.e. +/-, delay/no-delay) of each photoreceptor differing between the 2 RGCs. As such, we do not think that users will come away thinking that each photoreceptor can only inform center or surround exclusively. Nonetheless, to ensure that this important point is not overlooked, we now explicitly make this point in the manuscript and the lesson plans. Furthermore, in the lesson plans we now summarize concepts that are explored with each specific activity.

Sixth, the reviewers point out that RetINaBox failed to capture ON vs. OFF visual processing. To address this gap, we have implemented ON and OFF visual processing into RetINaBox. We have added this functionality to the software by allowing users to toggle RGCs between ON and OFF types within the Connectivity Manager. Accordingly, we have also added a figure to the manuscript (new Figure 2), and added a section about ON and OFF processing to Lessons 1 and 2. Furthermore, in the Lesson 1 lesson plan and teaching slides we now outline ON and OFF processing, including details about how the real retina implements sign- conserving OFF and sign-inverting ON responses via OFF and ON bipolar cells. Thus, RetINaBox now directly models how "every photoreceptor [can] contribute to both 'on' and 'of' channels... and to participate in the 'center' and 'surround' of different retinal ganglion cells." What educational level this tool would be appropriate for? In their closing sentence, the authors assert that it would be useful for students in high school to graduate school. In the current configuration it seems it may being useful for younger students, but graduate students would likely find it overly simplistic.

We specifically designed RetINaBox to be used for neuroscience outreach events, whether it be with students (high school or undergrad) or the general public. Regarding graduate students, while someone who is specifically studying circuit processing in the retina or visual cortex may find RetINaBox to be overly simplistic, we have found that it is engaging for graduate students working in molecular, genetic, and clinical aspects of vision, who often do not have backgrounds in circuits and systems neuroscience. We now more clearly make these points in the manuscript.

One important component of retina function that typically raises students' engagement comes from freely available movies showing that a large spot of light fails to elicit much response from the 'on' RGC, which is further deepened when the students see the somewhat puzzling suppression by the annulus of light. This provides a great opportunity to then delve into the RF structure and the underlying circuitry. The typical reaction to the movie of the 'end stopped' cell is even better. The point here is that these movies are freely available, and may do a better job of conveying the excitement of discovery than does RetINaBox.

Different people learn in different ways and we are not arguing that RetINaBox is the only way to engage with students. In fact, our lab is also working on more 'traditional' neuroscience learning tools (e.g. see www.theopenbrain.org). Furthermore, while we agree that videos are also great learning tools, we find it difficult to definitively state that one teaching method is "better" than another, as it likely depends on the specific student and the exact topic. We developed RetINaBox as we felt that a new learning tool that was focused on hands-on user- driven exploration and discovery could be a useful instrument to add to the existing repertoire of pedagogic tools, and be particularly engaging in outreach settings. We now more clearly outline this in the discussion.

The reviewers mention end stopped cells. We thank the reviewers for pointing out this gap. We have added an activity on end stopping to the orientation selectivity lesson (lesson 2).

It would be helpful to note whether the design could be scaled (e.g., by using GPIO expanders or multiplexing) to support more complex receptive field demonstrations. Even if the authors do not intend to pursue this now, acknowledging scalability as a possibility could broaden interest and highlight the flexibility of the platform.

We now outline that there are a few remaining GPIO ports available if users want to use them for additional purposes. We also now better emphasize how the software is completely open-source, and so if users are interested in expanding or scaling the functionality of RetINaBox, this is possible.

Minor issues.

What is the actual estimated cost and the amount of time necessary to make the device. From looking over the user's manual, it seemed that it would be a fairly time-consuming task. It would be helpful if the authors included a general cost estimate and some idea of how long it might take, say, a high school science teacher to build.

In the methods we now provide the cost (< $350 USD, including a Raspberry Pi and a monitor) and the time it takes to build RetINaBox (~4hrs, once you have the 3D printed parts and electronics components in hand).

There were also a couple of minor things that raises concerns, such as the statement that center-surround RFs of different sizes were for detecting objects of different sizes. Most vision scientists think of center-surround as a mechanism for reducing redundancy and signaling contrast. Different center sizes are important for conveying information about different spatial frequencies, but "object" is a much more complex and nuanced computation. Another was the question setting up lesson 3: "How do we know if and how objects are moving relative to us?" The kind of simple motion detector being built up with pure excitation and phase delays is a far cry from the circuits that detect relative motion.

We thank the reviewers for these notes. Throughout the manuscript, lesson plans, and teaching slides, for each of the 4 concepts that we explore (ON/OFF, center-surround, orientation selectivity, and direction selectivity) we now more clearly provide background info, example biological implementations, and ethological implications.

Specifically related to these reviewer comments:

First, we have reworded the sections of the lesson plan and teaching slides where we discussed objects, to make sure that what we are saying is both technically correct while also not being too technical for people first being exposed to these topics (in short, we removed the word "object" and now speak more generally of visual stimuli).

Second, regarding the role of center-surround receptive fields, we now more clearly outline that receptive fields relate to the part of the visual field a neuron 'sees', and that center- surround has two main functions: a) making neurons selectively respond to luminance contrast within their RF (meaning most visual neurons do not respond well to homogeneous (spatially redundant) images); b) the size of the center-surround RF dictates the spatial frequency of luminance contrast that a neuron is selective for.

Third, regarding differentiating between different types of motion, we agree with the statement above, but felt it was important to point out different ethological uses for direction selectivity. We have kept the point about differentiating self-generated vs. externally- generated motion since it's an ethological reason for motion detection, but we now more clearly state in the lesson plan that RetINaBox provides only a simple version of motion detection.