Topologically controlled circuits of human iPSC-derived neurons for electrophysiology recordings

Bottom-up neuroscience, which consists of building and studying controlled networks of neurons in vitro, is a promising method to investigate information processing at the neuronal level. However, in vitro studies tend to use cells of animal origin rather than human neurons, leading to conclusions that might not be generalizable to humans and limiting the possibilities for relevant studies on neurological disorders. Here we present a method to build arrays of topologically controlled circuits of human induced pluripotent stem cell (iPSC)-derived neurons. The circuits consist of 4 to 50 neurons with well-defined connections, confined by microfabricated polydimethylsiloxane (PDMS) membranes. Such circuits were characterized using optical imaging and microelectrode arrays (MEAs), suggesting the formation of functional connections between the neurons of a circuit. Electrophysiology recordings were performed on circuits of human iPSC-derived neurons for at least 4.5 months. We believe that the capacity to build small and controlled circuits of human iPSC-derived neurons holds great promise to better understand the fundamental principles of information processing and storing in the brain.


Survival rates
To estimate the survival rate, iNeurons were seeded into two sets of samples containing PDMS microstructures, as detailed in Section 2.5.1 of the main text. One of the sets of samples was supplemented with 1 μg/mL of laminin in the first week of culture ("Laminin" condition) while the other was not ("No laminin" condition). The samples were stained at DIV 0 and restained at DIV 11. The images were cropped into four individual images of nodes (N = 240 nodes per condition).

Image analysis
Fig. S1 shows representative images of a fluorescently labelled circuit of iNeurons. Such images were used to obtain the plots shown in Fig. 2. The number of live cells on DIV 0 and DIV 11 images were analysed according to the method presented in section 2.5.1 of the main paper. The number of dead cells at DIV 0 was estimated by processing the red channel of the image of each node. The red channel image was smoothed using a mean filter with a radius of one pixel. Default thresholding was then used to obtain a binary mask 1 , followed by the built-in Fiji Watershed algorithm to separate neighboring particles. Finally, the number of particles was counted for each mask. Fig. S2a and b show a step-by-step example of the automatized counting of live and dead cells for an example node for both the "no laminin" and "laminin" conditions. Fig. S2c compares the numbers obtained from the images shown in (a) and (b) with the manual hand count. Automatized counting is within 15% of the manual count. Fig. S2d shows the manual count of the live cells at DIV 11. By DIV 11, dead cells were not uniformly stained and some of them had separated into several pieces, making the use of segmentation unreliable to count the number of dead cells.

Survival rate at DIV 0
Survival rates at DIV 0 were calculated for both sets of samples. The survival rate of node i at DIV 0 (r 0,i ) was calculated for each node as: r 0,i = n live,0,i (n live,0,i + n dead,0,i ) with n live,0,i the number of live cells and n dead,0,i the number of dead cells at DIV 0 for node i. The resulting survival rates at DIV 0 for both sets of samples can be seen in Fig. S3. Importantly, laminin was added to the medium of the second set of samples at the same time as the CMFDA/ethidium homodimer-1 stains, 1 h after seeding. In the samples where no laminin was added, the average survival rate was 73%, whereas in the samples where laminin was added, the survival rate at DIV 0 was 67%. The null hypothesis of the statistical test that we used to compare both sets of samples was that there was no difference in the distribution of the percentage of live cells between the two samples at DIV 0. The survival rate was significantly higher in the samples without laminin vs. the samples where laminin was added. As laminin is not expected to have an instant effect on the samples, the differences observed in the initial survival between the two sets of samples is not attributable to the addition of laminin, but likely due to stochastic variations in the number and survival of cells contained in the volume pipetted onto the sample during the initial cell seeding.

Absolute survival rate at DIV 11
Fig. 2c presents the survival rate at DIV 11 relative to the number of cells alive at DIV 0. Based on the number of live and dead cells estimated from the DIV 0 and DIV 11 images, it is also possible to calculate the absolute survival rate at DIV 11, i.e. the survival rate relative to the total number of cells seeded at DIV 0. The absolute survival rate of node i at DIV 11 (r 11,i ) can be calculated for each node as: 1 Electronic Supplementary Material (ESI) for Lab on a Chip. This journal is © The Royal Society of Chemistry 2022 r abs,11,i = n live,11,i (n live,0,i + n dead,0,i ) with n live,11,i the number of live cells at DIV 11 for node i. Here we assume that the total number of cells per node did not vary between DIV 0 and DIV 11 and used the total cell count from DIV 0 for each respective node to calculate the absolute survival rate. This was possible because the same circuits were stained at DIV 0 and restained at DIV 11. Estimating the number of dead cells from DIV 11 images was impossible, as by that time, many of the dead cells had degraded into several pieces and were overlapping, making it difficult to get a reliable count of dead cells from the red channel images (see Fig. S2d).
The resulting absolute survival for both the "laminin" and the "no laminin" conditions can be seen on Fig. S4. For the "no laminin" samples, the average survival is of 0.4%, while it is of 2.9% for the "laminin" samples, a significant difference. In this case, the difference in survival at DIV 11 can be attributed to the addition of laminin in the medium. Fig. S5 shows representative images of fluorescent labelled circuits of iNeurons at DIV 4 and 7. Such images are the ones that were used to obtain the plots shown on Fig. 3. Each image of a full circuit was cropped into its four nodes. The area of the node occupied by green-and red-stained structures at the different DIVs was measured according to the method presented in section 2.5.2 of the main text. Fig. S6 shows examples of the steps used to obtain the binary masks from which the area occupied by live and dead cells in the node was calculated. The percentage area was obtained by dividing the white area by the area of the node (2.27x10 − 2 mm 2 ).

Open cultures of iNeurons
To investigate survival in the absence of PDMS microstructures and determine if PDMS is the leading cause of the low survival rate of iNeurons, iNeurons were plated on bare PDL-coated glass at a high density (300 k cells/cm 2 ) and the change in area occupied by live and dead structures over night was investigated.

Substrate preparation and staining
A glass bottom 48-well plate (P48G-1.5-6-F, Mattek) was used as a substrate. It was plasma cleaned for 2 min and coated with 0.1 mg/mL PDL in PBS for 45 min before being rinsed 3 times with PBS and left in ultrapure water. iNeurons were seeded at a density of 300 k cells/cm 2 on 18 of the wells of the glass bottom well plates. Two different conditions were tested: culturing the samples with regular medium (9 wells) and with medium containing 1 μg/mL of laminin (9 wells). At DIV 0, 1, 2, and 3, two wells of each condition were stained with CMFDA and ethidium homodimer-1 and 15 to 25 fields of view were taken for each well. At DIV 10, one well of each condition was stained with the live/dead and Hoechst stains and 16 fields of view were imaged. The other 16 wells were restained to investigate the effect of early staining on cell survival.
The area occupied by green-and red-stained structures were calculated using image segmentation, as detailed in Section 2.5.2 of the main text. To calculate the percentage of the area occupied by green-or red-stained structures, the measured area (in μm 2 ) was divided by the total area of a field of view. Fig. S7 shows examples of the steps used to obtain the binary masks from which the area occupied by live and dead cells was calculated. Due to the microscopy settings used, fields of view from DIV 0 were smaller than images from the other DIVs (424 µm vs 626 µm of side). This was taken into account by calculating the percentage of the field of view occupied by live-or red-stained structures.  Fig. S9 shows representative images of fluorescently labelled iNeurons at DIV 1, 2 and 3. From DIV 2, live iNeurons tended to cluster and overlap, making it difficult to reliably count the number of cells per field of view. For that reason, the area of the field of view occupied by green-or red-stained structures was used as a proxy for investigating the change over time of live and dead neurons. The area occupied by live and dead cells was hypothesized to correlate with the number of live and dead cells. Fig. S8c and d show the area occupied by green-and red-stained structures over DIV 0, 1, 2, 3 and 10 for the "no laminin" and the "laminin" samples. A statistical test was run to test the difference in the distribution of both sets of samples. Because laminin was added at the same time as the stains on DIV 0, which is not expected to have an immediate effect on survival, the null hypothesis for the statistical test on DIV 0 was: there is no difference in the distribution of the percentage of live cells in the two initial sets of wells, consisting in thawed iNeurons. On DIV 1 to 10, laminin had been added to the samples for more than a day. The null hypothesis for these experiments was therefore: there is no difference in the distribution of the percentage of live cells in the samples supplemented with 1 μg/mL of laminin for a week ("Laminin") vs. in the samples that were not supplemented with laminin ("No laminin").

Change in the area occupied by live and dead iNeurons over time
The area occupied by live iNeurons decreases by a factor of 4 in the first two days of cultures, before gradually decreasing until DIV 10 (Fig. S8c). Adding laminin to the medium does not have a measurable impact on the area occupied by live cells at DIV 10. There is a statistically significant difference in the area occupied by live cells at DIV 0 between the "no laminin" and "laminin" conditions. This difference is likely due to stochastic differences in the number of cells initially pipetted onto the substrates.
Based on the red-stained area measurements, the number of dead cells is fairly constant over the first three days, before slowly increasing between DIV 3 and DIV 10 (Fig. S8d). The area occupied by dead cells is on average higher in the "laminin" than in the "no laminin" condition at DIV 1, 3 and 10, with a statistically significant difference at DIV 1 and 3. This could be explained by the fact that laminin tends to make the surface of the substrate slightly more cell-adhesive than bare PDL, leading to more dead cells adhering to it rather than getting washed away during medium changes. Overall, even in an open culture surface at a rather high cell density, many of the iNeurons die over time, especially in the first few days of culture.

Effect of CMFDA/Ethidium homodimer-1 staining in early days of culture
We observed that staining cultures at early DIV has an adverse effect on the cell survival at later DIVs. We could quantify that effect by restaining at DIV 10 the wells of the 48-well plate used to obtain the plots shown on Fig. S8. Results of the area occupied by live structures on restained and non-restained wells can bee seen on Fig. S10. Compared to wells that were stained for the first time at DIV 10, restaining wells that had already been stained at DIV 0, 1, 2 or 3 resulted in 1 to 1.5 times lower area occupied by live cells. Early stains thus seem to have an adverse effect on survival and should be avoided. They can be replaced with genetically expressed fluorescent proteins. However, if using a typical cytosolic fluorescent proteins such as GFP, dead cells might also have expressed the fluorescent protein before dying, leading to unclean fluorescent images that usually cannot be automatically segmented and analyzed.

PDMS treatment
To determine if part of the cell death could be attributed to poorly treated PDMS, the effect of different PDMS cleaning methods prior to substrate making was assessed. The cleaning methods tested were autoclaving, solvent extraction and ethanol (A15, Thommen-Furler AG). Autoclaving consisted in placing the PDMS micostructures in an autoclave (Varioklav 75T, Sterico) and heating them at 121°C and 110 kPa for 20 min, followed by a 20 min drying cycle at a temperature of 81°C to 91°C. Solvent extraction was performed according to the extraction protocol reported by Millet et al. 2 . Ethanol cleaning consisted in immersing the PDMS membrane in ethanol for approximately 16 h, followed by 24 h of drying in an oven at 60°C. Fig. S11 shows the distribution of the number of nodes with at least one live iNeuron for the different PDMS cleaning methods. It seems like the different cleaning methods had either little effect (autoclaving) or an adverse effect on the number of full circuits (solvent extraction and ethanol), so none of these PDMS treatments were kept in the substrate preparation protocol.

Macrophage co-culture
Because of the poor survival rate, dead iNeurons accumulate in the nodes of the PDMS circuits and might have an adverse effect on the remaining live iNeurons in the node. To test for this, iPSC-derived macrophages were added to the cultures on DIV 4. iPSC-derived macrophages were obtained following the protocol described by Giorgetti et al. 3 and kindly provided by Novartis as a suspension.
Upon reception, macrophages were centrifuged for 5 min at 1000 rpm, resuspended in macrophage medium and plated into a non-coated 6-well plate (92006, TPP). Macrophage medium consisted in RPMI 1640 GlutaMax (61870-036, Ther-moFisher) with 10% heat inactivated FBS (A156-152, ThermoFisher), 1% sodium pyruvate (11360-039, ThermoFisher), 1% pen-strep (15140, ThermoFisher), 50 µM of mercaptoethanol (31350-010, ThermoFisher) and 40 ng/mL of human M-CSF (216-MC-025, Biotechne). Macrophages were kept in incubator for 8 days. They were then detached using TrypLE (12604-013) and centrifuged at 500 rpm for 5 min. Macrophages were seeded on top of a DIV 4 culture of iNeurons in PDMS microstructures at a density of about 50 k cells/cm 2 . Phase contrast images of the wells were taken every other day until DIV 20. At DIV 20, a live/dead and Hoechst stain was performed on them. Fig. S12a shows an example of one node getting cleaned by the macrophages over time. Impressively, the macrophages seem to phagocytose all of the dead cells contained in the nodes. Macrophages can squeeze into the low part of the chamber (see DIV 8 image) and move around the chamber a lot. Fig. S13 shows a comparison between macrophage-containing and no macrophage-containing cultures. In the presence of macrophages, live iNeurons tend to tightly cluster together and their axons are grouped into fairly straight bundle connecting all four nodes (see Fig. S12b and Fig. S13b). This is likely due to macrophages moving around the nodes and leading to a mechanical bundling of the axons along the most straight path connecting one node to the next. However, the presence of macrophages does not seem to lead to a big improvement on the number of full nodes. Adding macrophages lead to a more complex protocol, along with adding uncertainties of mixing several cell types whose in vivo interactions are not fully understood. For those reasons, co-culture of macrophages and iNeurons was not further explored. Fig. S14 shows representative examples of circuits with four nodes containing at least one live iNeuron for all of the conditions listed in Fig. 4b. A stitched image of all of the 15 circuits from one PDMS membrane can be seen on Fig. S15. This was obtained with condition 7 (10µg/mL of laminin).

Statistical tests on protocol variations
The number of live iNeurons per circuit after 3 weeks (DIV 18 to 23) was counted for all of the conditions presented in Fig. 4b. The statistical significance of the difference of the number of cells per circuit was tested by running pairwise two-sided Mann Whitney U tests on each pair of conditions. Results can be seen in Table S1.

Directionality
The "stomach" design of the PDMS microstructure allows to guide axons in a clockwise directions in 90% of the cases. This is due to the shape of the chamber and to the properties of the axons. When an axon starts growing from a single soma (Fig S16a), it grows until it hits a wall, then tends to follow the wall. If it grows towards the clockwise direction, it can simply follow the output channel to the next node (Fig S16b and c). If the axon grows towards the counter-clockwise direction, it will in most cases be redirected to the side channel, either because the axon is already following the top wall and naturally continues in the side channel (Fig S16d) or because it cannot follow the sharp angle from the input channel ( Fig S16e). In 10% of the cases, the axon manages to follow the sharp angle from the input channel ( Fig S16f) and connects in the wrong direction. All of these different possible cases were observed in circuits where few neurons survived. Examples for each case are shown in Fig S17, for both the 100-µm (top) and 170-µm (bottom) diameter node design. Fig S17f shows examples where several neurons were growing in the node.

Action potential waveforms
The action potential waveforms for all four electrodes of the circuit shown on Fig. 6 were extracted at DIV 36 and DIV 133. They are visible on Fig. S18.   4 Fig. S19 shows an overlay of the spikes detected from the four electrodes of a circuit during 60 s of spontaneous activity. The activity recordings come from the iNeuron circuit shown in Fig. 6 at DIV 21, 62, 90 and 133. Electrodes are colorcoded as in Fig. 6. To fit 60 s of spikes in one plot, each second of recording is visualized as two stacks of 500 ms, for a total of 120 stacks for each DIV. Some sequence of spikes repeat themselves, such as the "red-blue-red" pattern visible at DIV 21 or the many "red-green" patterns visible at DIV 62. Some more complex patterns are also visible at DIV 62 and 90. DIV 133 presents fewer of these patterns. The presence of these patterns of spikes suggests a temporal dependence between some of the electrodes of the circuit, a possible sign of functional connectivity. This was further inspected in Fig.  8 and 9.

Average number of active electrode per circuit
The average number of active electrodes per circuit of four electrodes was plotted in Fig. S20a. This plot shows a very similar trend than Fig. 7a: the addition of laminin to the medium leads to a much higher percentage of active electrodes.
The average MFR over all electrodes is shown in Fig. S20b. Since more than half of the electrodes of the "no laminin" MEAs are inactive, the blue MFR is substantially reduced, giving a false impression that the presence of laminin increases the firing frequency of the circuits, when in fact it is not the case. The presence of laminin simply increases the number of circuits with live cells and hence the number of circuits firing, but it does not increase the firing rate of the live cells.  Fig. S22 shows the full 500 ms post-stimulus response for the circuit shown in Fig. 9d. Bands corresponding to the temporally consistent response to the stimulus are mainly concentrated in the first 20 ms of recordings, so a zoom-in into the first 20 ms of the response can be seen in Fig. 9e.