Hindbrain V2a neurons impose rhythmic activity on motor neurons in an in vitro reticulospinal circuit

The reticulospinal system is an evolutionarily conserved pathway among vertebrates that relays locomotor control signals from the hindbrain to the spinal cord. Recent studies have identified specific hindbrain cell types that participate in this circuit, including Chx10+ neurons of the medullary reticular formation, which project to the spinal cord and are active during periods of locomotion. To create a system in which reticulospinal neurons communicate with spinal motor effectors, we have constructed an in vitro model using two purified excitatory neuronal subtypes: HB9+ spinal motor neurons and Chx10+ hindbrain neurons. Cultured separately, these neurons exhibit cell type-specific patterns of activity; the Chx10+ cultures developed regular, synchronized bursts of activity that recruited neurons across the entire culture, whereas motor neuron activity consisted of an irregular pattern. A combination of the two subtypes produced cultures in which Chx10+ neurons recruited the motor neurons into synchronized network bursts, which were dependent on AMPA receptors. In addition to demonstrating that the activity of in vitro networks can depend on the developmental identity of their constituent neurons, we provide a new model with genetically specified nerve cell types to study the activity of a reticulospinal circuit. Significance statement Models of the brain that use cultured neurons are usually comprised of a complex mixture of different kinds of cells, making it hard to determine how each cell type contributes to the overall pattern of activity. We made a simplified culture containing two cell types known to form a reticulospinal circuit in vivo. While in isolated cultures, each cell type had a distinct pattern of activity, in coculture the activity of one cell type came to dominate, indicating that the patterns observed in complex neuronal cultures arise in part from the distinctive properties of the constituent neurons.

of their constituent neurons, we provide a new model with genetically specified nerve cell types 23 to study the activity of a reticulospinal circuit. 24

Significance statement: 26
Models of the brain that use cultured neurons are usually comprised of a complex 27 mixture of different kinds of cells, making it hard to determine how each cell type contributes 28 to the overall pattern of activity. We made a simplified culture containing two cell types known 29 to form a reticulospinal circuit in vivo. While in isolated cultures, each cell type had a distinct 30 pattern of activity, in coculture the activity of one cell type came to dominate, indicating that 31 the patterns observed in complex neuronal cultures arise in part from the distinctive properties 32 of the constituent neurons. 33

Introduction: 34
To derive clinically relevant findings, most in vitro models of neurological disease seek to 35 incorporate as realistic a mixture of cells from the modeled region as possible. These models of 36 disease hold therapeutic promise, as some have already been used to identify small molecule 37 candidates for drug development (Yang 2013 properties of individual neuronal cell types, it is important to study each cell type in isolation 50 prior to combining them into a more complex system. 51 To address this question, we created a simple system from two neuronal subtypes with 52 a well-defined relationship in the intact nervous system, motor neurons and reticulospinal 53 neurons. Motor neurons relay patterned input from spinal cord central pattern generator 54 circuits to skeletal muscles to initiate behavior (Binder 1996). Reticulospinal neurons are 55 components of a prominent behavioral circuit that relays rhythmic locomotor drive from the 56 brainstem to the spinal cord (Peterson 1979, Garcia-Rill 1987a,b , Le Ray 2011, Kiehn 2016). 57 7 embryos. Sorted Chx10 + hindbrain neurons were seeded on either 5mm glass coverslips in a 24-136 well plate or multi-electrode arrays, both prepared with a confluent layer of glia, at a density of 137 1x10 4 neurons/well of coverslips or 4x10 4 neurons/array. All Chx10 + hindbrain neurons were 138 cultured in Neurobasal medium supplemented with 2% SB-27 (Gibco, 17504044), 1% GlutaMax, 139 1% pen/strep/antimycotic, 1µM dbCaMP, 10ng/ml BDNF, 10ng/ml GDNF, and 1µM ascorbic 140 acid. 141 For reticulospinal cocultures, sorted HB9::GFP + motor neurons and Chx10::CFP + 142 hindbrain neurons were seeded together on a confluent layer of glia on either 5mm coverslips 143 or multi-electrode arrays. On coverslips in a 24-well plate, HB9 + neurons were seeded at a 144 density of 2.5x10 5 cells/well and Chx10 + neurons were seeded at a density of 1x10 5 cells/well. 145 On multi-electrode arrays, HB9 + neurons were seeded at a density of 1x10 6 cells/dish and 146 Chx10 + neurons were seeded at a density of 4x10 5 cells/dish. Cocultures were grown in the 147 same supplemented BrainPhys medium used for HB9 + cultures. 148

Animals 149
Mice were group housed in a 12-hour light/dark schedule, with food and water provided 150 ad libitum. For timed matings, two females were introduced into the home cage of a single 151 male, where they remained for the duration of the mating. Females were checked for vaginal 152 plugs every 24 hours and removed to separate cages after plug was detected, and singly housed 153 for the duration of the timed pregnancy. All animal procedures and protocols were approved by 154 bring Vm to -70mV and current steps were applied in 10pA increments from -10 to 130pA for 1 203 second duration, returning to -70mV holding potential between steps. For voltage clamp 204 experiments, the cell was held at -80pA for 100ms before stepping voltage injection from -100 205 to 150mV in 10mV increments for 100ms, returning to the -80pA holding potential between 206 each step. 207 Data analysis and plotting of patch clamp data were performed using ClampFit and 208 Matlab (see github.com/abubnys/patch_clamp_analysis for specific scripts used). To generate 209 IV plots of voltage-gated sodium current from voltage-clamp data, the local minimum evoked 210 current within 30ms of voltage step onset was subtracted from the mean current during the 211 last 30ms of the voltage step and plotted against the magnitude of the injected voltage. 212

Multi-electrode recordings 213
Multi-electrode arrays were cultured with HB9 + motor neurons or Chx10 + hindbrain 214 neurons as described above (see Cell Culture methods). For the duration of the lifetime of the 215 culture (D3 to D30 days after plating for Chx10 + and D7 to D30 days after plating for HB9 + 216 neurons), spontaneous extracellular activity was recorded using the MEA2100-Lite system 217 (MultiChannel Systems). The array was placed in the recording apparatus and allowed to 218 equilibrate at room temperature for 30 minutes prior to recording for 4 minutes. Data 219 acquisition was performed on MCRack with an input voltage range of -19.5 to +19.5mV and a 220 sampling frequency of 20kHz. Raw electrode data for 60 electrodes were processed through a Bessel 4 th order high pass filter with a cutoff at 400Hz. The spike detection threshold was 5 222 standard deviations below the mean of the filtered recordings. Raw and filtered data, along 223 with spike timestamps were converted to .txt files using MC_DataTool and the resulting files 224 were analyzed in Matlab. 225 video, brightness and contrast was adjusted uniformly across the image stack using the "auto" 306 adjust function. Then, the minimum intensity for each pixel across the image stack (calculated using the "Z-project, minimum" function) was subtracted from each image in the stack to 308 remove noise. Then, brightness and contrast were adjusted again across the image stack.  To isolate pure populations of HB9 + spinal motor neurons and hindbrain Chx10 + 348 neurons, we employed fluorescence activated cell sorting (FACS). We cultured these cell types 349 as single populations and also as a mixed reticulospinal culture. We differentiated HB9 + spinal 350 motor neurons from HBG3 embryonic stem cells using Wichterle et al's protocol to induce the spinal motor neuron identity (Wichterle 2002). Embryoid bodies were dissociated 6 days after 352 formation and sorted on the basis of HB9::GFP expression. E14 stem cells lacking GFP were 353 used as a negative control for FACS (Figure 1b,c). Approximately 50-60% of unsorted cells in the 354 embryoid body derived from HBG3 ES cells expressed GFP. FACS sorting for GFP expression 355 enriched this population to >96% purity. HB9::GFP + motor neurons were subsequently cultured 356 on a layer of cortical astrocytes to improve axonal outgrowth and network development. 357 To test whether sorting affected the electrophysiological activity of HB9 + neurons, we 358 performed whole cell patch clamp on HB9::GFP + neurons from sorted and unsorted cultures 359 grown in parallel under identical conditions. After 7 days in culture, HB9 + neurons in both 360 treatments responded to brief current pulses with spike trains, having a spike threshold around 361 20pA (Figure 2a,b). They developed voltage gated sodium current (INa) with maximum current 362 evoked at -2±12 mV (Figure 2c-e) that was not significantly different between sorted and 363 unsorted populations (Student's 2-tailed T-test p = 0.879). After 13 days in culture, both sorted 364 and unsorted HB9 + motor neurons also developed spontaneous spike trains (Figure 2f). 365 We then isolated and cultured primary hindbrain neurons expressing the transcription 366 factor Chx10, also using the FACS approach. We first assessed the Chx10 + neurons' behavior in 367 vitro as a homogeneous population, and then in combination with HB9 + neurons to determine if 368 they could form a reticulospinal circuit in vitro. For these experiments, we dissected neurons 369 from embryonic Chx10::CFP +/mice at E12.5, prepared a single cell suspension and used FACS to 370 isolate the CFP + population. As a negative control for CFP expression, we used hindbrains taken 371 from wildtype (WT) Swiss Webster E12.5 mouse embryos that do not express CFP (Figure 1d,e). 372 The hindbrains contained 2-3% Chx10::CFP + neurons, and sorting enriched this population to 375 >95% purity. These CFP + neurons were then cultured on a layer of cortical astrocytes, which is 376 known to improve the development and long-term viability of neuronal cultures (Wang 1994, 377 Maher 1999, Boehler 2004. 378 It is possible that, when removed from the intact reticular formation with its descending 379 inputs and diversity of other cell types, Chx10 + hindbrain neurons would not develop any 380 intrinsic activity that could pattern a reticulospinal circuit. To assess the electrophysiological 381 development of sorted Chx10 + neurons, we used whole-cell patch clamp to record the 382 spontaneous activity of single cells in cultures at different ages ranging from 1 to 30 days in 383 culture. For Chx10 + hindbrain neurons, the measured membrane capacitance was 22.75±2pF, 384 membrane resistance was 787.27±105MW, access resistance was 29.01±3MW, and membrane 385 voltage was -22.6±4mV. We found that Chx10 + hindbrain neurons developed spontaneous 386 electrophysiological activity after 5 days in culture. This activity started off as random trains of 387 spikes, but gradually became organized into robust, regular bursts by 10 days in culture and this 388 pattern of activity continued throughout the remaining lifetime of the cultures (Figure 2g). 389

Motor and Chx10 neuron cultures develop distinct patterns of network activity 390
Having established that HB9 + motor neurons and Chx10 + hindbrain neurons develop 391 spontaneous electrophysiological activity at the single cell level, we sought to determine 392 whether cultures of either cell type, which are composed almost exclusively of excitatory 393 neurons and astrocytes, could generate spontaneous patterns of network activity, whether 394 these patterns would organize into network bursts, and whether there were any cell-type 395 specific differences in such activity. To record the activity of multiple neurons at different time points, we cultured sorted 399 HB9 + motor neurons on multielectrode arrays (MEAs) containing a grid of 64 extracellular 400 recording electrodes. We recorded their spontaneous activity daily over 30 days, starting from 401 the day after plating. 402 We found that on their own, without astrocytes, sorted HB9 + motor neurons did not 403 develop any spontaneous activity on the MEA (n = 6). However, when these neurons were 404 cultured on a confluent layer of astrocytes, they gradually developed robust network activity 405 that remained stable over a month of recording (n = 14). We note that astrocytes cultured on 406 their own did not develop spontaneous activity when recorded on MEAs (n = 3), although we 407 did observe spontaneous calcium flux in astrocyte cultures visualized with the calcium-sensitive 408 dye Rhodamine3 (Video 3-1). The activity of HB9 + motor neuron/astrocyte cultures was not 409 well coordinated, even among neighboring recording electrodes (Figure 3a-c). To assess 410 whether the overall activity of the culture had a hidden underlying temporal structure, we 411 calculated the mean spike rate across all active channels of the HB9 + motor neuron cultures and 412 found that it remained constant throughout the recording session (Figure 3d). 413 When we used the calcium-sensitive dye Rhodamine3 to assess HB9 + motor neuron 414 activity with single-cell resolution, we observed randomly distributed calcium spikes that were 415 asynchronous between neighboring neurons (Figure 3e, Video 3-2), though more mature 416 cultures did develop some synchrony (Figure 3f, Video 3-3). The mean correlation coefficient 417 between the spike rates of multiple neurons within the same HB9 + neuron culture was 418 0.15±0.17 (p = 0.15). 419

Figure 3
When we cultured Chx10 + hindbrain neurons on MEAs with a confluent layer of 423 astrocytes, we observed the emergence of spontaneous activity with these neurons as well. 424 Unlike HB9 + neurons, Chx10 + neurons developed robust and coordinated network bursts 425 (Figure 3g-j). Practically no spikes occurred outside of these sharply delineated bursting 426 periods. The time between bursts (inter-burst interval) varied between 2 and 10 seconds 427 throughout the lifetime of the cultures, with no apparent long-term trend. We observed the 428 same sort of robust network bursts in Chx10 + hindbrain neuron cultures with calcium imaging 429 (Figure 3k, Video 3-4). 430

Chx10 + neurons impose their activity patterns on HB9 + neurons in coculture 431
Despite their common glutamatergic identity, we observed that HB9 + and Chx10 + 432 hindbrain neurons develop distinct patterns of spontaneous network activity. If these two cell 433 types fail to form functional connections to one another in vitro, these patterns of activity 434 should remain unchanged in coculture, but if a unidirectional functional connection forms 435 between Chx10 + and HB9 + neurons, we might expect to see one activity pattern dominate in 436 coculture. To test these possibilities, we cultured the two cell types together as a mixed 437 population on MEAs and recorded their spontaneous activity daily over 30 days. 438  Corresponding right panels indicate identified neurons quantified for calcium activity, white arrowheads for Chx10::CFP + and yellow arrowheads for HB9::GFP + .

Such cocultures develop spontaneous bursts of comparable time scale and duration to pure 441
Chx10 + cultures, though some neurons continue to have spiking activity that resembles HB9 + 442 motor neurons in between network bursts (Figure 4a-c). When the overall network activity was 443 measured by averaging spike rates across all active electrodes, the Chx10-like network bursts 444 predominated (Figure 4d). 445 It is possible that the bursts we observed in the reticulospinal culture were generated 446 only by the Chx10 + neurons in the dish and that the HB9 + motor neurons were quiescent and 447 did not contribute to network activity. In order to determine which cell type participates in the 448 cultures' network bursts, we used calcium imaging to obtain single cell resolution recordings of 449 the coculture. We found that neighboring HB9 + and Chx10 + neurons both participate in network 450 burst events (Figure 4e, Video 4-1). Some HB9 + motor neurons in coculture also have brief, 451 non-coordinated calcium spiking events that occur between the larger bursts (Figure 4f, Video  452   4-2). 453 The percentages of Chx10 + and HB9 + neurons from calcium-imaging experiments that 454 were spiking, bursting, both spiking and bursting, or inactive in each of the culture conditions 455 are summarized in Table 1

HB9 + and Chx10 network activity is an AMPA receptor-dependent process 459
The spontaneous coordinated activity we observed in Chx10 + and HB9 + neuron cultures 460 could be the product of intrinsic pacemaker properties of these neurons or an emergent 461 property of the network that is dependent on synaptic transmission. To distinguish between 462 these alternatives, we applied a panel of synaptic blockers targeting α-amino-3-hydroxy-5- We repeated the CNQX drug application on cocultures and used calcium imaging with 484 Rhodamine3 to visualize the activity of the culture prior to and after application of 40µM CNQX. 485 Despite a loss of network bursting activity, we observed that some HB9 + neurons in the 486 coculture continued to have spontaneous spiking activity in the presence of a blocking 487 concentration of CNQX (Figure 5h-j). 488 We also tested the effects of the NMDA receptor antagonist AP5 on all three cultures 489 (Figure 6a-f 4). H-J, Calcium imaging of co-culture H, bursting prior to CNQX application (shown also in video 5-1), I, inhibition of bursting, but not Hb9::GFP + spiking, by application of 40 µM CNQX (shown also in video 5-2), and J, bursting recovers after washout of CNQX (shown also in video 5-3). Corresponding right panels are photomicrographs of neurons quantified for calcium activity, indicated by white arrowheads for Chx10::CFP + and yellow arrowheads for HB9::GFP + . 496 497 Figure 6 Discussion: 499 In this study, we used flow cytometry to isolate HB9 + motor neurons and Chx10 + 500 hindbrain neurons and cultured these cell types separately and together to form a 501 reticulospinal circuit. We found that the sorting process did not significantly impact the 502 development of HB9 + and Chx10 + neuron electrophysiology. When isolated, these two cell 503 types developed distinct patterns of network activity. HB9 + neurons tended towards 504 uncoordinated spike trains, while Chx10 + hindbrain neurons were characterized by regular, 505 network-wide bursts of activity. Cocultures of these two cell types developed the network 506 bursts characteristic of Chx10 + neurons that recruited neighboring HB9 + neurons. We further 507 note that the activity of all these cultures was insensitive to NMDA and GABAA receptor 508 blockers but could be inhibited by the AMPA receptor blocker CNQX. Our observation that HB9 + motor neurons fail to develop spontaneous activity in the 531 absence of glia is consistent with other studies that have demonstrated the essential support 532 that astrocytes provide for cultured neurons (Wang 1994, Boehler 2007, including motor 533 neurons (Ullian 2004). When we cultured sorted HB9 + neurons with astrocytes they developed 534 unsynchronized spike trains. FACS appears to be critical for this behavior, as previous studies of 535 stem cell-derived HB9 + motor neurons cultured without FACS isolation reported coordinated network bursts (Jenkinson 2017). In unsorted cultures, a cell type other than HB9 + neurons 537 must have contributed to the generation of this activity pattern. We note that the spinal motor 538 neuron differentiation protocol generates a small but prominent subpopulation of spinal V2a 539 neurons, a cell type that is closely related to our rhythmogenic hindbrain Chx10 + neurons 540 (Brown 2014). 541 We found that Chx10 + hindbrain neurons isolated by FACS and cultured on MEAs 542 developed robust and highly coordinated network bursts. Calcium imaging (Figure 3) indicates 543 that virtually all Chx10 + neurons participate in these bursts, with no discernible time delay. 544 Thus, simultaneous spiking is an intrinsic feature of Chx10 + neurons in culture and does not 545 appear to require the presence of other cell types. 546

Chx10-like pattern of activity is dominant in coculture 547
Recordings from the coculture indicate that Chx10 + neurons impose their rhythmic 548 bursting phenotype on adjacent HB9 + neurons (Figure 4e). In this way, we were able to induce 549 HB9 + neurons to participate in network bursts by exposing them to another rhythmogenic cell 550 type. Purified cultured HB9 + neurons did not coordinate their spontaneous calcium spikes 551 despite being adjacent to each other. Thus, HB9 + neurons appear to require patterned input 552 from another cell type. In contrast, Chx10 + hindbrain neurons are able to generate their own 553 patterns of activity without the need for exogenous cell types besides astrocytes. In summary, 554 our results indicate that when cultured with astrocytes, electrically excitable cell types develop 555 different spontaneous patterns of activity that are driven by the intrinsic properties of that cell 556 type. 557 We were unable to detect any spontaneous activity in cultures without astrocytes, 558 consistent with previous results with neuronal culture on MEAs (Ullian 2004, Boehler 2007. It is 559 likely that one way that astrocytes mediate such an effect is by removing excess glutamate to 560 prevent excitotoxicity (Rothstein 1996, Swanson 1997. Consistent with previous reports 561 (Scemes 2006), the astrocytes in our culture were active, as indicated by slow waves of calcium 562 activity which we were able to observe in calcium imaging (Video 3-1), but which did not 563 produce electrical excitation on MEAs. 564 Our observation that Chx10 + neurons are able to impose temporally patterned activity 565 on HB9 + neurons is consistent with their in vivo function of driving rhythmic behaviors such as 566 hindlimb locomotion and respiration. Prior studies suggest that activation of these neurons is 567 associated with bouts of locomotion, and may drive locomotor stop signals (Bretzner, 2013, 568 Bouvier 2015. Additionally, Chx10 + neurons project to the pre-Bötzinger complex, and their 569 ablation disrupts respiratory rhythms in newborn mice, with normal respiratory rhythms 570 gradually reasserting themselves as the mice grow older (Crone 2012, Crone 2009). 571

Emergent properties of neuronal cultures as revealed by synaptic inhibition 572
Our results from applying a panel of synaptic blockers targeting AMPA, NMDA, and 573 GABAA receptors to spontaneously active HB9 + and Chx10 + neuron cultures (Figures 5 and 6) 574 show that the AMPAR blocker CNQX effectively blocked all bursts in Chx10 + cultures and 575 significantly decreased the activity in HB9 + neuron cultures. This is consistent with the 576 observation that spinal motor neurons cultured in vitro form glutamatergic synapses that are 577 entirely blocked by CNQX (Ullian 2004). CNQX application similarly eradicates spontaneous 578 network bursting in cultures of spinal Chx10 + neurons that are otherwise insensitive to glycine and GABA antagonists (Sternfeld 2017). Our finding that bursts of hindbrain Chx10 + neurons 580 could be effectively eradicated by blocking glutamatergic transmission suggests that the robust 581 rhythmicity of these neurons is an emergent property of the network, as opposed to pacemaker 582 activity generated by individual cells. This contrasts with true pacemaker neurons, such as 583 those of the pre-Bötzinger complex, where bursts are intrinsic to individual cells, and therefore 584 insensitive to the same cocktail of synaptic blockers (Chevalier 2016). Thus, we observe that 585 AMPA receptor activation can drive very different outcomes that depend on cell type. 586 When we applied CNQX to the coculture, some neurons switched from rhythmic 587 bursting to a transient period of tonic spiking before becoming quiescent. This emergent 588 property may be driven by HB9 + neurons that revert to their native spiking phenotype in the 589 absence of the patterning influence of network bursts. This is consistent with our calcium 590 imaging data in which we identified HB9 + neurons in coculture that continued to have calcium 591 spikes even in the presence of a dose of CNQX that effectively disrupted network bursts (Figure  592   5). 593

Implications of our results for modeling reticulospinal circuits 594
The results of our study could be applied to modeling of reticulospinal circuits, different 595 aspects of which are currently being examined by multiple groups (Sternfeld 2017, Oueghlani 596 2018, Pivetta 2014). In the rodent reticulospinal circuit, hindbrain Chx10 + neurons primarily 597 contact premotor networks within the spinal cord, as opposed to synapsing directly onto motor 598 neurons (Bouvier 2015). By contrast, in the zebrafish hindbrain Chx10 + neurons directly contact 599 spinal motor neurons and generate swimming when selectively stimulated (Kimura 2013). 600 input directly to motor neurons, driving sensory-evoked swimming before other motor control 602 systems have developed (Soffe 2009, Li 2019). Thus, it can be argued that the circuit created by 603 our in vitro cocultures replicates the basic circuitry found in fish and amphibians. It would be 604 interesting to determine whether the emergent properties of Chx10 + neurons from these 605 species differ from the mouse, and how incorporating additional reticulospinal cell types would 606 alter patterns of activity. 607 Ultimately, the most generalizable aspect of the findings we report here is the 608 observation that the aggregate activity of neuronal networks is influenced by the specific 609 molecular identity of their constituent neurons, beyond specific pacemaker cells or broad 610 categories of excitatory-inhibitory cells. Our results indicate that many electrical properties of 611 neurons are intrinsic to their specific subtype, which is an important consideration for modeling 612 the effects of mutations and disease on network function. The cell type compositions of circuit 613 models can have profound effects on patterns of activity and need to be considered and 614 interpreted carefully. 615