A disinhibitory nigra-parafascicular pathway amplifies seizure in temporal lobe epilepsy

The precise circuit of the substantia nigra pars reticulata (SNr) involved in temporal lobe epilepsy (TLE) is still unclear. Here we found that optogenetic or chemogenetic activation of SNr parvalbumin+ (PV) GABAergic neurons amplifies seizure activities in kindling- and kainic acid-induced TLE models, whereas selective inhibition of these neurons alleviates seizure activities. The severity of seizures is bidirectionally regulated by optogenetic manipulation of SNr PV fibers projecting to the parafascicular nucleus (PF). Electrophysiology combined with rabies virus-assisted circuit mapping shows that SNr PV neurons directly project to and functionally inhibit posterior PF GABAergic neurons. Activity of these neurons also regulates seizure activity. Collectively, our results reveal that a long-range SNr-PF disinhibitory circuit participates in regulating seizure in TLE and inactivation of this circuit can alleviate severity of epileptic seizures. These findings provide a better understanding of pathological changes from a circuit perspective and suggest a possibility to precisely control epilepsy.

1. The work of the late Dr. Karen Gale, who originally described the anti-seizure effects of nigral inhibition in the early 1980s, is not cited at all in this manuscript. Dr. Gale's name is synonymous with the substantia nigra within the field of epilepsy, and this oversight should be corrected. 2. The authors describe the SN-PF pathway as "previously unknown" which is surprising given this pathway was reported in 1979 by Beckstead, Nauta and colleagues (Brain Res, 1979). There is also no discussion of the role of the indirect SN-nRT-PF pathway originally described by Tsumori et al. (Brain Res, 2000). This is of particular interest as the authors found no effect of RT activation/inhibition in the current study. 3. There is no discussion of subregion within the SNpr that is involved. Work from the Redgrave and Moshe groups have both shown that not all regions of the SNpr are equivalent in their impact on seizure activity. Indeed, these groups have reported rostrocaudal differences, and have even suggested that the pars lateralis of the SN may in fact be the critical site. Along these lines there is also no discussion of prior work showing that activation of SNpr (pharmacologically) can increase seizure severity (see for example, Sperber et al., Brain Res, 1989). 4. Nail-Boucherie and colleagues (Epilepsia, 2005), reported that an excitatory projection from superior colliculus to PF is also involved in the control of seizures (the GAERS model of absence epilepsy). Given that the SC is a primary target of SNpr outflow, there appears to be convergence in the circuit. This may merit some discussion. 5. In 1987, Garant and Gale reported that ascending projections from the SNpr to the thalamus were not necessary for the anticonvulsant action of nigral inhibition against maximal electroshock seizures. While the seizures in the MES model are quite different from those in the kindling model or the intrahippocampal KA model, this may also merit some discussion. 6. Methodological Question -for the single unit analysis, it is not described how excitation, inhibition and no response were defined, nor is it clear the number of trials (seizures) recorded for each unit. Did the proportion of neurons displaying an excitatory response exceed that to be expected by chance? 7. Single Unit Analysis -If multiple seizures were recorded within subject, did the response change across repeated stimulation? Bonhaus et al., (J Neurosci, 1986) previously reported striking differences in the engagement of SNpr neurons in naïve and kindled animals. This should be clarified. Note that this same question applies to the photometry data. 8. Photometry -it has become more or less standard in the field to include imaging at a calciuminsensitive excitation wavelength for GCaMP (or imaging in GFP controls) to ensure that responses are not due to issues such as movement artifact. Indeed, even small deflections in the fiber optic cable can produce large amplitude responses during photometry. Can the authors provide evidence that the photometry signal they report is biologically relevant and not due to movement artifact? 9. The use of parametric statistics for the kindling progression is slightly concerning, as these data are inherently non-normal (they are categorical). Perhaps the authors might consider using an approach like the Aligned Rank Transform for nonparametric ANOVA to address this (see Wobbrock et al for details). 10. Why do the authors break out Stage 2, Stage 4 and Fully Kindled, as compared to any of the other kindling stages in terms of number of stimulations to reach these thresholds? 11. For Figs 1 and 2, there is appreciable variability in the response to photostimulation within the treated groups, it would be particularly interesting to know how the optic placement and expression of the animals varied in this case. More generally, it is unclear if/how optic placement and opsin expression was verified for each subject. 12. Fig 3 -DREADD experiment in chronic TLE -the manuscript would be strengthened by including appropriate controls for Fig 3D-H (i.e., DREADD negative animals treated with CNO) as the authors acknowledge that CNO can be associated with off target effects. 13. Fig 4 -(1) what effect does BIC in PF have on seizures in the absence of light stimulation? (2) how do the authors account for potential antidromic activation of SNpr-PF terminals, and subsequent activation of other nigral targets? 14. Clarification for Figure S5 -for Panels C-I, how many mice were used for these experiments (I am assuming the individual data points represent cells, not animals?) 15. The finding that the neurons that receive input from SNr in the PF thalamus are local circuit is quite interesting -and somewhat surprising given that others (see: Bentivoglio et al., Exp Brain Res 1991; Arcelli et al., Brain Res Bull, 1997) have argued that interneurons are sparse or absent in the intralaminar thalamus of rodents, a profile that differs from that in the cat and primate. The author should discuss this point -are they certain these neurons didn't project elsewhere? 16. Fig S7 would benefit from DREADD-negative controls. 17. Minor comment: The axons of transfected neurons have been reported to remain photoexcitable even when severed from their parent somata 35, 36. This line on Pg 14 is out of context. Reviewer #2: Remarks to the Author: The authors present a commendable enormous amount of work to decipher the SNr components that may control seizures in TLE. They use a combination of techniques to propose that a GABAergic circuit involving the SNr inhibition of PF neurons is central to the control of seizure propagation. The role of the SNr in TLE has been proposed a long time ago; authors using various protocols such as DBS, lesions, and injections of chemical. Note that the authors forgot to mention the seminal work of Karen Gale. The concept is not novel, but the detailed mechanism was not identified. I have a major concern. The authors claim that "These findings provide a better understanding of the pathological network changes and the precise spatiotemporal control of epilepsy." However, 90% of the "TLE" experiments are done in control animals, which are kindled. This type of acute model can be used to propose testable hypotheses to chronic models. However, only one set of experiments is performed in a chronic model of TLE. As a result, from kindled animals, it is not possible to claim "a better understanding of the pathological network changes" because there are no major network alterations in the progressive kindling model. Likewise, it is not possible to conclude that we get a better understanding of "the precise spatiotemporal control of epilepsy". In TLE (both in patients and chronic models), there are major anatomical and functional alterations in SNr and thalamic nuclei. The circuits are reorganized, and we do not know whether results obtained in control animals can be generalized to chronic model. There is nothing wrong with doing a study using the kindling model, but the impact of the conclusions is limited, by the very nature of the approach. My evaluation is that relying mostly on kindled normal animals to draw important conclusions on TLE is not enough for Nature Comm. As it stands, the study belongs more to a specialized journal. My evaluation would be entirely different if the authors were to 1. The authors claim that SNr PV cells control TLE seizures via a new "disinhibitory" projection from SNrPV cells to GABAergic cells in Pf, rather than via the direct projection of SNr to glutamatergic cells. "Quantity analysis showed the number of SNr PV neurons targeting PF glutamatergic neurons was much lower than that targeting PF GABAergic neurons (Fig. S6)." However, SNrPV stimulation inhibits a non-negligible population of Pf GLU cells as well (see Fig 5B). In fact, SNr stimulation inhibits the same number of GLU and GABA cells in Pf according to In other words, in light of the results shown in Fig. 5B and Figs. S5 and S6, the overall working model is unclear. The proportion of GABA vs GLU cells in Pf is only 5% in the Pf nucleus (see Fig S5A); Moreover less than 20% of these GABA Pf cells receive a functional input from SNr ( Fig S5, patching results: only 3 GABA cells out of 17 receive SNr inputs). Could this low number be due to sub-optimal recording conditions? What is the rationale for using 4-AP and TTX when recording evoked IPSCs from SNr to Pf thalamus?
Can the authors speculate on how the SNr projection onto just 1% of Pf nucleus is more important for seizure control than the direct targeting of GLU Pf cells by SNr (which would represent a higher number of cells according to Fig 5B, although inconsistent with Fig S6)?
In Fig S7, the authors nicely show that the chemogenetic activation of PF GABAergic neurons blocks pro-epileptic effects of optogenetic activation of SNr-PF GABAergic projections. However, I am not convinced that this result can be interpreted as if the pro-epileptic effect of SNr activation was due 100% to GABA Pf cells and did not involve SNr projections to GLU Pf cells (because see Fig 5B, there are GLU Pf cells that seem to be inhibited by SNr). Thus, the fact that the pro-epileptic effect of SNr activation was reduced in the Fig S7 experiment could be due to a combined effect of a direct inhibition of Pf GLU cells by SNr AND to inhibition of Pf GLU cells by chemogenetic activation of GABA Pf cells.
The authors do claim that GLU Pf neurons are important for SNr-mediated modulation of TLE seizures (Fig S9 and last part of the discussion), but it is unclear how the authors reconcile the direct SNr effects on GABA vs GLU Pf cells. Fig. 5: The authors conclude that the anti-epileptic effect of optogenetic activation of PF GABAergic neurons was reversed by intra-PF application of the GABAA receptor antagonist bicuculline. However, bicuculline could affect both glutamatergic and GABAergic cells in PF, both of which receive GABAergic inputs from SNr. This caveat should be addressed in the text because it affects data interpretation.
In sum, it would significantly strengthen the manuscript if the authors could propose a working model  / diagram based on their results. The authors should edit the text to make it very clear what would  one would expect from activation or inhibition of SNrPV cells on seizures with and without existence of  GABAergic Pf cells (even if these are present in relatively small numbers). How does SNr PV projection compare between GABAergic and glutamatergic Pf cell population? This is key for speculating about the relevance of the "new" circuit. In other words, the presence of GABAergic cells in Pf is novel in this manuscript, so the authors should clarify with a concluding/speculative diagram how the presence of these cells and the fact they receive projections from SNr would actually affects the effect of SNr on TLE seizures.
2. Given that the authors claim that the output of Pf cells is important for controlling seizures (which is the major claim of the study), I recommend showing the firing of these cells during seizures in the main figures rather than in the Supplemental figure 9.
3. Given that the existence of GABAergic cells in Pf thalamus is novel (because the Pf neurons are thought to be only glutamatergic), it is important to characterize the properties of these cells. Please note that TLE is not an epileptic seizure. TLE is a form of epilepsy. This sentence needs to be rewritten. Suggestion: "Basal ganglia circuits are closely involved in the modulation, propagation, and cessation of seizures in different types of epilepsies, including TLE 7, 8." -Also, the references 7 and 8 do not encompass the broad claim about basal ganglia controlling several types of seizures. Basal ganglia have also been shown to control absence type epileptic seizures (see work from the Charpier group on how SNr projections control thalamocortical oscillations in epilepsy, e.g., PMID:17251435). The authors cite the references correctly in the discussion but not in the introduction. b) Page 3, Introduction, The reference cited in this sentence is incorrect: "The substantia nigra pars reticulata (SNr), a region that mainly contains GABAergic neurons, controls the activity of both corticothalamic and limbic networks through its primary GABAergic output, acting like the "choke point" of basal ganglia 9". Reference 9 corresponds to: "Paxinos G. The rat nervous system. (ed^(eds). Fourth edition. edn. Elsevier/Academic Press, (2015)" and is not the right reference for the notion of choke points. For the notion of choke points, the authors should cite relevant manuscripts (e.g the reference 2 from this manuscript which is the one that focuses on the notion of choke points). c) Intro page 4: What neurons do the authors refer to by "these neurons" in the following section: "We identified a previously unknown nigra-parafascicular disinhibitory circuit for regulation of seizure in TLE. We found that selective activation of these neurons amplifies seizure activities, whereas inhibition of SNr PV neurons alleviated the severity of epileptic seizures". Do they mean the following: "We found that selective activation of SNr PV neurons amplifies seizure activities, whereas their inhibition alleviated the severity of epileptic seizures" ? d) In the same paragraph, the authors write: ".
[…] the SNr PV neurons sent long-range axons to the GABAergic neurons in posterior parafascicular nucleus (PF), and optogenetic manipulation of this disinhibitory nigra-parafascicular circuit bidirectionally modulated seizures in TLE." This statement could be interpreted as if SNr PV neurons do not project to glutamatergic neurons in Pf that are known to be the major cell type in Pf. However, SNr PV projections to Pf do not seem to be cell-type specific (according to the text) because SNr PV cells seem to be projecting to Pf glutamatergic cells as well (e.g Fig 5B), suggesting that the direct SNr->GLU Pf pathway needs to be considered when interpreting the results from  Indicate in the panels, or remove from the legend. Also, Po is indicated in panels, but the abbreviation is not defined in the legend.
-Panel C: the authors claim in the legend and in the main text that SNr projects to RT but I can't see any projections from SNr to RT in the panels. If the authors want to report this projection, I suggest adding a zoom on the RT to show presence of SNr fibers/synaptic boutons. h) Fig S1: legend: In the sentence: "…ipsilateral SNr from 3 wildtype mice": the authors should clarify ipsilateral to what? i) There are many grammatical and typing mistakes throughout the text. It would be too long to list all of them. E.g., "These results indicate that activation of SNr GABAergic neurons is sufficient to amplifies seizure activities in hippocampal kindling model".
Reviewer #4: Remarks to the Author: In this study Chen et al analyzed the involvement of the substantia nigra pars reticulata (SNr) in temporal lobe epilepsy (TLE) by employing the kindling and the kainic acid animal models. They report that optogenetic or chemogenetic activation of SNr parvalbumin+ (PV) interneurons "amplifies" seizures activities while their inhibition alleviates them. They also state that seizure severity was bidirectionally regulated by optogenetic manipulation of SNr PV fibers projecting to the parafascicular nucleus (PF), and found with electrophysiology combined with rabies virus-assisted circuit-mapping that SNr PV neurons directly project to and functionally inhibit posterior PF GABAergic neurons. They conclude that their findings reveal "a previously unknown long-range SNr-PF disinhibitory circuit that modulate seizures in TLE, and that inactivation of this circuit can alleviate epileptic seizures. This is a complex study that includes a variety of experimental approaches providing a huge amount of data. The topic is relatively unexplored and the findings original and consistent with the conclusion that the SNr-PF circuit modulates seizures. However, at this stage, this study is characterized by a rather superficial analysis of the results, which may be due to the amount of different experiments performed. Perhaps the authors should focus on some specific experiments and provide full, in depth information on them.
Some specific critiques are as follows.
1. In the main body of the paper it should be specified whether the "CA3 kindling" was acute or the classic chronic kindling. This point is of paramount importance to follow what written in the Results (p. 5), namely "Immediately after the initiation of CA3 kindling, about 55% of GABAergic neurons in the ipsilateral SNr increased their firing rate (peak firing rate was 17.06 ± 1.01 Hz; out of 38 recorded neurons from 3 wildtype mice, 21 neurons were excited, 4 inhibited, 13 no response; no inter-animal statistical difference was detected, Chi-square test, p=0.8927, Fig. S1B and F1C)." In fact, it is obscure to me whether these changes occurred during the kindling stimulation or after several days of kindling.
2. The same critique applies to what stated in p. 6: In the hippocampal kindling model, we applied 30 s photostimulation in the ipsilateral SNr immediately after hippocampal kindling…" More specific information on the timing of optogenetic stimulation in relation with the kindling procedure must be given.
3. I have some concerns with the "negative" data obtained by stimulating SNr NOS positive neurons (see p. 8), which according to these authors is the second largest population of GABAergic cells in the SNr. Did they check for expression? How come, if effectively activated by optogenetic procedures these interneurons did not influence the "kindling" process?? In fact, were they optogenetically excited?? 4. As remarked above, with regard to the kindling experiments, also the chemogenetic procedures applied to kainic acid treated mice (pp. 10 and 11) lack experimental details. 5. There is actually some debate in the literature regarding the role of CNO in chemogenetics. Some studies have shown that CNO does not cross the blood-brain barrier but that is instead converted to clozapine, which acts as a D1R agonist. Since it is known that the activation of D1R modulates the release of GABA from SN terminals, it would be interesting to see whether a similar effect on chronic seizures in the KA model is also observed in the PVCre + CNO group ( Figure 3). 6. Page 6: It is stated that "Representative peri-event histograms confirmed that blue-light stimulation (473 nm, 20 Hz, 10 ms, 5 mW, 10 s on-off cycle) excited 15 out of 19 SNr GABAergic neurons from 3 Vgat-ChR2-eYFP mice, suggesting that SNr GABAergic neurons can be functionally activated by bluelight stimulation. Yellow-light stimulation (589 nm, continuous light, 5 mW, 10s on-off cycle), serving as the control stimulation, had no effect on the same neurons (Fig. S1F). It is unclear why a 20 Hz blue-light stimulation was used to activate SNr GABAergic neurons whereas continuous yellow light stimulation was used for control experiments. A similar comment would apply to the results shown in figure 1 G-1 if the same procedure was used. 7. Page 9: "The number of stimulations needed to reach stage 2 was not affected here (only has a tendency, U=19, P=0.1740)". I would suggest removing "only has a tendency" since the p value of 0.17 is not close to significance. 8. Page 29: It is unclear whether the status epilepticus induced with local administration of KA was pharmacologically stopped or allowed to self-terminate. Status epilepticus of different durations could lead to differences in seizure occurrence and neuropathology during the chronic period. Please specify.
9. Page 30: It is stated that "Spontaneous seizure events were defined as regular spike clusters with a duration of ≥10 s, spike frequency of ≥2 Hz and amplitude at least three times of the baseline EEG, accompanying behavioral tonic-clonic GSs". Does this mean that non-convulsive seizures that were only visible on the EEG were not considered?
The paper should also be carefully edited. A few examples -p.2, in the Introduction "…remote brain regions, extending from cortical to subcortical limbic structures and other remote structures." -p.3 in the Introduction: "…targeting at the SNr…" -p. 5 in the Results: "in the previous study." -p.19: "Furthermore, optogengetic activation of PF glutamatergic neurons was sufficient to accelerate seizure development in hippocampal kindling model".

The work of the late Dr. Karen Gale, who originally described the anti-seizure effects of nigral inhibition in the early 1980s, is not cited at all in this manuscript. Dr. Gale's name is synonymous with the substantia nigra within the field of epilepsy, and this oversight should be corrected.
Reply: Thank you very much for this comment. We have included the work of Dr. Karen Gale in our revised manuscript.

Page 3-4 in revised manuscript:
--"Experimental studies have also reported structural and functional changes of SNr neurons among different types of epileptic models 14, 15, 16, 17, 18 . In addition to those findings, lesion 19 , pharmacological interference 20, 21, 22, 23 or deep brain stimulation 24, 25 targeting the SNr can regulate the process of epileptic seizures, suggesting that the SNr plays a key role in seizure control."--

Page 22 in revised manuscript:
--"Interestingly, SNr GABAergic neurons control absence epilepsy through the downstream VM thalamocortical circuit 9 , while nigrothalamic/nigrostriatal projections do not appear to contribute to the electroshock convulsions 51 , suggesting that basal ganglia play a role in various seizure disorders through distinct downstream circuits."--New added references: Reply: Thank you very much for this insightful comment. Indeed, not all subregions of the SNr may be equivalent in their impact on seizure activity, as we can see appreciable variability in the response to optogenetic modulation in Fig.1 and 2. Here, we did not further analysis how the fiber location within the SNr would have a correlation with the treated effects, because apart from fiber location, there are other factors (the amount of viral expression, area size of illumination, et.al.) that would affect seizure-modulating effects. We do agree with the reviewer that precise location of SNr subregion is an important subject for future precise seizure control, as previous studies have indicated that there are rostro-caudal differences of SNr in seizure control. Accordingly, we have cited corresponding papers and added some discussions in the revised manuscript as following:

Page 3-4 in revised manuscript:
--"In addition to those findings, lesion 19 , pharmacological interference 20, 21, 22, 23, 24 or deep brain stimulation 25, 26 targeting the SNr can regulate the process of epileptic seizures, suggesting that the SNr plays a key role in seizure control."--

Page 21 in revised manuscript:
--"It should be noted that not all regions of the SNr may be equivalent in their impact on seizure activity, as we can see appreciable variability in the response to optogenetic modulation in Fig.1  48. Shehab S, Simkins M, Dean P, Redgrave P. Regional distribution of the anticonvulsant and behavioural effects of muscimol injected into the substantia nigra of rats. The European journal of neuroscience 8, 749-757 (1996). , 2005), reported that an excitatory projection from superior colliculus to PF is also involved in the control of seizures (the GAERS model of absence epilepsy). Given that the SC is a primary target of SNpr outflow, there appears to be convergence in the circuit. This may merit some discussion.

Nail-Boucherie and colleagues (Epilepsia
Reply: Thank you very much for this comment. Accordingly, we have added some discussions in the revised manuscript as following: Page 21-22 in revised manuscript: --"As a common issue in optogenetic and chemogenetic modulation, targeting hub regions, such as the SNr or PF here, may alter global brain dynamics. Therefore, apart from SNr-PF circuit, other potential global alterations may also be account for the factors for the regulator seizure in TLE. For example, the superior colliculus, a primary target of SNr output, sends glutamatergic projections to the PF, which was previously reported to be involved in the control of absence seizures 50 . It is possible that the superior colliculus may also have a potential role in seizure of TLE."-- Reply: Thank you very much for this comment. Accordingly, we have added some discussions in the revised manuscript as following:

Page 22 in revised manuscript:
--"Interestingly, SNr GABAergic neurons control absence epilepsy through the downstream VM thalamocortical circuit 9 , while nigrothalamic/nigrostriatal projections do not appear to contribute to the electroshock convulsions 51 , suggesting that basal ganglia play a role in various seizure disorders through distinct downstream circuits."--New added reference: 51. Garant DS, Gale K. Substantia nigra-mediated anticonvulsant actions: role of nigral output pathways. Experimental neurology 97, 143-159 (1987).

Methodological Question -for the single unit analysis, it is not described how excitation, inhibition and no response were defined, nor is it clear the number of trials (seizures) recorded for each unit. Did the proportion of neurons displaying an excitatory response exceed that to be expected by chance?
Reply: Thank you very much for this comment. As previous study 1 , the criteria used to define an "excited" or "inhibited" neuronal response was as follows: firing rates were considered to be significantly different if they were >2 SDs of baseline averages. Briefly, the average firing rate during each 10-s-duration bin was calculated in 1-min baseline period. Then, the average firing rate of the seizure period or photostimulation was calculated and compared with that of the baseline period to test whether firing rate was >2 SDs greater or less than the baseline average. In single-unit recording experiments with optogenetic modulation, we found blue-light stimulation excited putative PV neurons (10 out of 22 neurons from 3 PV-ChR2 SNr mice), while yellow-light stimulation, as a control light, had no effect on the same neurons (Fig. 1E), suggesting that the proportion of neurons displaying an excitatory response may not be expected by chance. To make it more clear, we have added detailed information in the revised method as following: Page 33-34 in revised manuscript: --"The criteria used to define an "excited" or "inhibited" neuronal response were as follows: firing rates were considered to be significantly different if they were >2 SDs greater or less than baseline averages 66 . Briefly, the average firing rate during each 10-s-duration bin was calculated in 1-min baseline period. Then, the average firing rate of the seizure period or photostimulation was calculated and compared with that of the baseline period to test whether firing rate was >2 SDs greater or less than the baseline average."-- Reply: Thank you very much for this important comment. For single-unit analysis, we did single-unit recording in anesthesia mice, so we did not measure the response change across repeated stimulationsinduced seizure develepment. While, for fiber photometry experiment, we recorded the Ca 2+ response change across repeated stimulation, we found that calcium signal of SNr PV neurons gradually increases during kindling acquisition (Fig. S2a). Combined with previous report from Bonhaus et al., this data suggested that basal firing rate of SNr GABAergic neurons and their response to seizure significantly increased after acquisition of generalized seizure. Thus, we have added this data into Fig. S2a as following:

Reference
Revised Fig. S2a 8. Photometry -it has become more or less standard in the field to include imaging at a calciuminsensitive excitation wavelength for GCaMP (or imaging in GFP controls) to ensure that responses are not due to issues such as movement artifact. Indeed, even small deflections in the fiber optic cable can produce large amplitude responses during photometry. Can the authors provide evidence that the photometry signal they report is biologically relevant and not due to movement artifact?
Reply: Thank you very much for this important suggestion. To test whether the increased calcium signal was due to movement artifact, we injected AAV-ef1a-DIO-eYFP control virus into the SNr of PV-cre mice (PV-eYFP SNr mice), and recording calcium response during kindling-induced seizure. We found that there is no change of calcium signal in PV-eYFP SNr mice, suggesting that calcium signal of SNr PV neurons was caused seizure activities, but not movement artifact. Therefore, we have included this data in the Fig. S2b. 9. The use of parametric statistics for the kindling progression is slightly concerning, as these data are inherently non-normal (they are categorical). Perhaps the authors might consider using an approach like the Aligned Rank Transform for nonparametric ANOVA to address this (see Wobbrock et al for details).
Reply: Thank you very much for this comment. Accordingly, we have used the Aligned Rank Transform for nonparametric ANOVA for the statistical test for the development of seizure stage during kindling progression in the revised manuscript as previous study 1 . Reply: Thank you very much for these insightful comments. Fluorescent expression of optogenetic or chemogenetic virus was examined for all mice with Olympus FV-1000 imaging system. If the optical fiber was located above (within 0.5 mm) the targeted region with correct viral expression, the corresponding mouse was taken into statistical analysis, since previous study indicated that surface positioning of the optical fiber allows effective stimulation with ~500 μm depth of cortical cortex. We have also noticed there is appreciable variability in optogenetic modulation experiment in Fig. 1  and 2 variability. Please see all the raw data in the "source data" file. We did not further analysis how the fiber location within the SNr would have a correlation with the treated effects, since apart from fiber location, there are other factors (the amount of viral expression, area size of illumination, et.al.) that would affect seizure-modulating effects. We do agree with the reviewer that precise location of SNr subregion is an important subject for future precise seizure control, as previous studies have indicated that there are rostrocaudal differences of SNr in seizure control. Accordingly, we have added corresponding discussion in the revised manuscript and more detailed description for method as following:

Reference
Page 36 in revised manuscript: --"Fluorescent expression of optogenetic or chemogenetic virus was examined for all mice. If the optical fiber was located above (within 0.5 mm) the targeted region with correct viral expression, the corresponding mouse was taken into statistical analysis"--

Page 21 in revised manuscript:
--"It should be noted that not all regions of the SNr may be equivalent in their impact on seizure activity, as we can see appreciable variability in the response to optogenetic modulation in Fig.1 and 2 48. Shehab S, Simkins M, Dean P, Redgrave P. Regional distribution of the anticonvulsant and behavioural effects of muscimol injected into the substantia nigra of rats. The European journal of neuroscience 8, 749-757 (1996).

Fig 3 -DREADD experiment in chronic TLE -the manuscript would be strengthened by including appropriate controls for Fig 3D-H (i.e., DREADD negative animals treated with CNO) as the authors acknowledge that CNO can be associated with off target effects.
Reply: Thank you very much for this constructive suggestion. To test whether the CNO itself would have any seizure-modifying effect in chronic epileptic mice, we injected AAV-ef1a-DIO-mCherry control virus into the SNr of PV-cre mice (PV-mCherry SNr mice). We found that CNO treatment did no change the frequency and seizure duration of both FS and GS in PV-mCherry SNr mice, suggesting anti-seizure effect of CNO in PV-hM4Di SNr mice may not be associated with off target effects of CNO. Thus, we have added this result in the revised Fig. 3G as following:

Revised Fig. 3G
Page 11 in revised manuscript: --"CNO itself did no change the frequency and seizure duration of both FS and GS in PV-mCherrySNr mice (Fig. 3G), suggesting anti-seizure effect of CNO in PV-hM4Di SNr mice may not be associated with off target effects of CNO."--

Fig 4 -(1) what effect does BIC in PF have on seizures in the absence of light stimulation? (2) how do the authors account for potential antidromic activation of SNpr-PF terminals, and subsequent activation of other nigral targets?
Reply: Thank you very much for these insightful comments.
(1) We found that bicuculline injected into the PF alone had no obvious effect on kindling acquisition (Please see Figure for reviewer 1 below). It can be caused by the non-specific effect of pharmacological modulation on PF neuron. As it was also referred by the third reviewer, bicuculline could affect both glutamatergic and GABAergic cells in PF, both of which receive GABAergic inputs from the SNr or even other regions. This caveat may complicate data interpretation. To avoid confusing, we did not include this data in the manuscript.
(2) To test whether there is a potential antidromic activation of SNr soma when activating SNpr-PF terminals, we performed c-fos immunohischemical experiment in the SNr. We found that there is no c-fos positive neuron after long-term photostimulation of SNpr-PF terminals. As a positive control for c-fos antibody, we found there is large amounts of c-fos positive staining in the motor cortex after seizure activities (Please see Figure for reviewer 2 below). Meanwhile, in our study, we found that optogenetic inhibition of SNpr-PF terminals, which may not produce potential antidromic effects on SNr soma, retarded seizure spread in kindling model (Fig. 4). Thus, these data suggested that PF might be the main downstream brain region that involved in seizure modulation of SNr PV neurons. All these cells are from 8 slices of 6 mice. We have corrected this mistake in the revised manuscript as following:

Page 8 in revised supplementary information:
--"The number of cells (from 8 slices of 6 mice) used in each group is indicated in figure."--

The finding that the neurons that receive input from SNr in the PF thalamus are local circuit is quite interesting -and somewhat surprising given that others (see: Bentivoglio et al., Exp Brain Res 1991; Arcelli et al., Brain Res Bull, 1997) have argued that interneurons are sparse or absent in the intralaminar thalamus of rodents, a profile that differs from that in the cat and primate. The author should discuss this point -are they certain these neurons didn't project elsewhere?
Reply: Thank you very much for this important comment. In our study, we found that the expression of Arch-eYFP or ChR2-mCherry was localized to the PF in Vgat::Arch or Vgat::ChR2 mice, respectively (Fig.  5 b and f), suggesting that GABAergic neurons are local, but not projecting neurons. Meanwhile, the antiepileptic effect of optogenetic activation of PF GABAergic neurons was reversed by intra-PF application of the GABAA receptor antagonist bicuculline (Fig. 5h), suggesting there is a local GABAergic microcircuit within the PF regulating the seizure propagation. To further verify that PF GABAergic neurons are local, we checked the Arch-eYFP expression in whole brain slices after injecting very little virus (50 nl) in the PF of Vgat-cre mice to prevent viral spillover and found no projecting terminals outside the PF. The typical pictures of viral expression are showed below (Figure for reviewer 3). Thus, all above data indicated that GABAergic neurons in the PF are local, but not projecting neurons.
As previous studies suggest that interneurons are sparse or even absent in the intralaminar thalamus of rodents, in the present study we first reveal the function of PF GABAergic neuron in epilepsy. Thus, we emphasized this novelty in the discussion of revised manuscript.

The authors present a commendable enormous amount of work to decipher the SNr components that may control seizures in TLE. They use a combination of techniques to propose that a GABAergic circuit involving the SNr inhibition of PF neurons is central to the control of seizure propagation. The role of the SNr in TLE has been proposed a long time ago; authors using various protocols such as DBS, lesions, and injections of chemical.
Note that the authors forgot to mention the seminal work of Karen Gale. The concept is not novel, but the detailed mechanism was not identified.
I have a major concern. The authors claim that "These findings provide a better understanding of the pathological network changes and the precise spatiotemporal control of epilepsy." However, 90% of the "TLE" experiments are done in control animals, which are kindled. This type of acute model can be used to propose testable hypotheses to chronic models. However, only one set of experiments is performed in a chronic model of TLE. As a result, from kindled animals, it is not possible to claim "a better understanding of the pathological network changes" because there are no major network alterations in the progressive kindling model. Likewise, it is not possible to conclude that we get a better understanding of "the precise spatiotemporal control of epilepsy". In TLE (both in patients and chronic models), there are major anatomical and functional alterations in SNr and thalamic nuclei. The circuits are reorganized, and we do not know whether results obtained in control animals can be generalized to chronic model.
There is nothing wrong with doing a study using the kindling model, but the impact of the conclusions is limited, by the very nature of the approach. My evaluation is that relying mostly on kindled normal animals to draw important conclusions on TLE is not enough for Nature Comm. As it stands, the study belongs more to a specialized journal. My evaluation would be entirely different if the authors were to replicate their main findings in the chronic model. They have tried to verify one aspect (but there are major issues there, which are discussed below), so they have the possibility to do it.
Reply: Thank you very much for reviewing our work and propounding valuable comments, which are helpful for us to improve the quality of our paper greatly. Notably, we have included the work of Dr. Karen Gale in our revised manuscript and have verified main findings (Fig.4-6) in chronic KA TLE model. Below are our point-by-point responses to the comments.

A major issue is the interpretation of most experimental results. The authors nicely show that a transition activation/inactivation of neurons/axons produce the expected effects in the recorded cells. But these experiments were performed under anesthesia with urethane that directly affects GABAergic transmission, thus acting as a confounding factor. In addition, in the kindling model, the optogenetic manipulation occurs for each stimulation. We do not know how cells will react to a buildup of activation (or Cl loading). The authors need to show that the activation/inactivation do work recording units in awake animals (there could be ceiling effects, or reversal, or post inhibitory rebound, which is typical for some PV cells).
Reply: Thank you very much for these important comments. We agree with the reviewer that anesthesia may affect GABAergic transmission and thus acting as a confounding factor for the data interpretation. Accordingly, we have performed additional unit recording experiments in awake mice.
First, we were aimed to tested whether optogenetic stimulation, using the similar parameter in kindling model, were able to reliably activate SNr PV neuron in awake mice. We verified that there is the increased firing of the SNr GABAergic neurons reliably in response to blue-light stimulation (473 nm, 20 Hz, 10 ms, 5 mW, "30s-on, 10s-off" cycle) in awake Vgat-ChR2-eYFP mice (Fig. S1G).
Furthermore, in KA-induced chronic awake PV-ChR2 SNr-PF mice, we found that optogenetic activation of SNr-PF PV projections can inhibit PF GABAergic neurons (2/4) and activated a part of PF glutamatergic neurons (10/24 from 3 mice) reliably during the whole photostimulation period (Fig. S7C-D). Interestingly, the modulating effect was much stronger in the initial seconds than that in later stage during the whole photostimulation period (Fig. S7D), suggesting that there might be accumulative Cl-loading in postsynaptic neurons that weaken optogenetic manipulation. Nevertheless, optogenetic activation of SNr-PF PV projections can still reliably inhibit PF GABAergic neurons during whole photostimulation period. After photostimulation, we did not detect any post inhibitory rebound firing. Thus, we have added these new data in the revised manuscript as following:

Revised Fig. S1G
Revised Fig. S7C-E 2. Characterization of seizures. Fig 1G and others. I do not understand. You indicated that you filtered the signal at 100Hz. Thus, you cannot see any frequency above 50 Hz at best, or 25 Hz. How did you perform the signal analysis and FFT? In addition, FFT is not appropriate for this type of dynamical signal, wavelet analysis is much better. In addition, you cannot calculate a power spectrum on a signal that is not in steady-state. Perhaps, you can look at seizure onset and offset. You may also consider using the coastline index to characterize seizures.
Reply: Thank you very much for this valuable comment. Actually, in our study, raw EEG signals were sampled at 1000 Hz and recorded with band-pass filters spanning DC to 200 Hz. Spectral analysis was conducted using the fast Fourier transform (FFT), which provided the total power between 0 and 100 Hz. We agree with the reviewer that FFT may not be appropriate for this type of dynamical signal, thus we used wavelet analysis to calculate the power spectrum of EEG during seizures (revised Figs.1G and 2E).
Meanwhile, as suggested by the reviewer, we also used coastline index to characterize seizures. We found that photo-activation of SNr PV neurons significantly increased the coastline index of EEG during seizures (revised Fig. 1G), while photo-inhibition of SNr PV neurons significantly decreased the coastline index of EEG during seizures (revised Fig. 2E). Thus, we have added this data in the revised manuscript as following: Revised Fig.1G Revised Fig.2E Page 31 in revised manuscript: --"Briefly, raw EEG signals were sampled at 1000 Hz and recorded with band-pass filters spanning DC to 200 Hz. Spectral analysis was conducted using the fast Fourier transform (FFT) or wavelet transform, which provided the total power between 0 and 100 Hz. The coastline index was determined as the sum of the absolute difference between successive points."-- Reply: Thanks very much for this comment. In acute KA model, behavior seizure stage accompanied by EEG activities are used to distinguish FS from GS. According to Racine scale, seizure stages 1-3 indicate FSs, and stages 4-5 are GSs. The seizure stage was scored by an investigator blinded to the group allocation.
In this acute seizure model, intra-hippocampal KA injection induced high amplitude spike-waves in the EEG and gradually elicited FS, GS and SE within tens of minutes. Seizure events were defined as regular spike clusters with a duration of ≥10 s, spike frequency of ≥2 Hz and amplitude at least three times of the baseline EEG. Once the mice were in a SE condition, they would still have several GSs during 1.5-h observation period. To make it more clear, we have showed the detailed LFP in three conditions and indicated the timing where GS onset (red arrowheads) in the revised Fig. 3C. Meanwhile, we have added detailed description in the revised method.

Revised Fig. 3C
Page 30 in revised manuscript: --"KA injection induced high amplitude spike-waves in the EEG and gradually elicited FS, GS and SE within tens of minutes. The seizure stage was scored by an investigator blinded to the group allocation during 1.5h observation period."--4. The latter leads to a major issue. Is it possible that CNO treatment and the heavy opto stimulations done during kindling change brain states globally? Targeting hub regions, such as SNr or thalamic nuclei may alter global brain dynamics. As a result, the effects may be due to a global alteration versus local one. This should be discussed at least.
Reply: Thank you very much for this insightful comment. Accordingly, we have added some discussion in the revised manuscript as following: Page 21-22 in revised manuscript: --"As a common issue in optogenetic and chemogenetic modulation, targeting hub regions, such as the SNr or PF here, may alter global brain dynamics. Therefore, apart from SNr-PF circuit, other potential global alterations may also be account for the factors for the regulator seizure in TLE. For example, the superior colliculus, a primary target of SNr output, sends glutamatergic projections to the PF, which was previously reported to be involved in the control of absence seizures 50 . It is possible that the superior colliculus may also have a potential role in seizure of TLE."--  First, we performed a morphological study to determine whether the amount of PV cells in SNr are changed in chronic epileptic model. Immunohistochemistry data showed that there is a minor decrease in the number of SNr PV neurons in KA-induced chronic awake mice (Fig. S7A, B). This is consistent with previous finding that there is neural loss in the SNr after animals have experienced epilepticus status 1,2 .

Chronic model. This is the model that should be used throughout. First, I advise performing a morphological study to determine the amount of cell loss (in particular PV cells) in
Second, in KA-induced chronic TLE model, spontaneous seizure events were defined as regular spike clusters with a duration of ≥10 s, spike frequency of ≥2 Hz and amplitude at least three times of the baseline EEG according to previous study 3 . In the present model, most of seizures we recorded were characterized by bursts of high-voltage sharp waves (revised Fig. 3E, upper panel) associated with any obvious behavioral alterations in the parallel video recordings. As previously reported 4 , those seizures only occurred in the ipsilateral hippocampus and usually were interpreted as non-convulsive focal seizures (FSs). In addition to FSs, mice also exhibited infrequent tonic-clonic generalized seizures (GSs) that were associated with typical paroxysmal EEG activity with obvious post-seizure depression (revised Fig. 3E, lower panel). To make it more clear, we have showed typical FS and GS in the revised Fig. 3E and added detailed description in the revised method.
Third, as also suggested by the first reviewer, to test whether the daily injection of CNO would have any seizure-modifying effect in chronic epileptic mice, we injected AAV-ef1a-DIO-mCherry control virus into the SNr of PV-cre mice (PV-mCherry SNr mice). We found that CNO treatment did no change the frequency and seizure duration of both FS and GS in PV-mCherry SNr mice, suggesting anti-seizure effect of CNO in PV-hM4Di SNr mice may not be associated with off target effects of CNO or daily injection manipulation. Thus, we have added this result in the revised Fig. 3G.
Finally, we have also noticed that the properties of FSs and GSs in HM3 and HM4 animals have a big variety. The precise reason is still unknown. We speculate this can be caused by the methodology of chronic KA model in our study. We stereotaxically injected KA into right dorsal hippocampus to induce SE in anesthetized mice. The mice were allowed to recovery from anesthetized without SE termination. This may lead to different SE duration in different mice, which may be an important contributor to the big variety among different mice. Thus, to make a fair comparation, we use self-control in chronic KA experiment.
Revised Fig. S7A --"First, we stereotaxically injected KA into right dorsal hippocampus to induce SE in anesthetized mice. The mice were allowed to self-terminate and recovery from anesthesia state."--

Page 30-31 in revised manuscript:
--"Most of seizures we recorded were characterized by bursts of high-voltage sharp waves and are not associated with any obvious behavioral alterations in the parallel video recordings, which usually were interpreted as non-convulsive FSs as previous study 65 . In addition to FSs, mice also exhibited infrequent tonic-clonic GSs that were associated with typical paroxysmal EEG activity with obvious post-seizure depression."-- In KA-induced chronic awake mice, although there is a minor reduce in the number of SNr PV neurons (Fig. S7A, B), we found that optogenetic activation of SNr-PF PV projections can also inhibit 2/4 putative GABAergic neurons and activated a part of putative glutamatergic neurons (10/24 from 3 mice) reliably during the whole photostimulation period (Fig. S7C-E). Furthermore, chemogenetic inactivation of SNr-PF PV projections by intra-PF injection of CNO, or chemogenetic activation of PF GABAergic neurons, similarly alleviated the severity of spontaneous seizure in KA-induced chronic epileptic model (Fig. S7F, G). Thus, all these above data suggested that SNr-PF neural circuit is involved in seizure control in KA-induced chronic epileptic model. --"To record optical evoked GABAergic postsynaptic currents, recording pipettes were routinely filled with a solution containing the following (in mM): 140 CsCl, 5 NaCl, 10 HEPES, 0.2 EGTA, 2 Mg-ATP, 0.3 Na3GTP, 5 QX314, 10 Na2-phosphocreatine, pH 7.2. Neurons were hold at -70 mV. The inhibitory postsynaptic currents (IPSCs) were isolated by addition of 6-cyano-7-nitroquinoxaline-2,3dione (CNQX; 20 μM) and (2R)-amino-5-phosphonovaleric acid (D-AP5; 20 μM) to the ACSF to block excitatory transmission mediated by AMPA/kainite and N-methyl-D-aspartate (NMDA) receptors, respectively. Experiments were performed in the presence of the sodium channel blocker tetrodotoxin (TTX, 1 μM) and the potassium channel blocker 4-amynopyridine (4-AP, 100 μM) to isolate the monosynaptic current while shining blue light on the surface of brain slices using an optic fiber connected to a blue laser power source (IKECOOL-Laser 473 nm). Light pulse (1 Hz, 10 ms, 10 pulses, 2 mW) was controlled with EPC10 patch-clamp amplifier (HEKA Instruments). To record the action potential firing properties of glutamatergic and GABAergic neurons in the PF, a series of hyperpolarizing and depolarizing current steps (-100 pA-+100 pA, 100 pA each a step, 500 ms) were applied to determine firing properties of PF neurons. A series of 300-ms depolarizing current pulses (increased in 5-pA steps) were applied to measure the threshold, amplitude, and the half-wave width of the action potentials. The pipette solution contained the following (in mM): 140 K-gluconate, 5 NaCl, 2 Mg-ATP, 0.2 EGTA, and 10 HEPES."--

Page 7 in revised supplementary information:
--"The number of cells (from 8 slices of 6 mice) used in each group is indicated in figure."--

Minor.
1. Fig S1C and elsewhere. With n=3 mice and few neurons, how did you assess inter-animal difference assessed? What were the criteria to decide that cells were excited/inhibited? What was the repartition per mouse?
Reply: Thank you very much for these comments.
The neuronal response repartition per mouse in Fig. S1C was showed in the below table. We used Chi-square test and found there is no inter-animal statistical differences (Chi-square=1.11, df=4, p=0.8927).
Meanwhile, as previous study 1 , the criteria used to define an "excited" or "inhibited" neuronal response were as follows: firing rates were considered to be significantly different if they were >2 SDs of baseline averages. Briefly, the average firing rate during each 10-s-duration bin was calculated in 1-min baseline period. Then, the average firing rate of the seizure period or photostimulation was calculated and compared with that of the baseline period to test whether firing rate was >2 SDs greater or less than the baseline average.
To make it more clear, we have included raw data in the "source data file" and added the criteria of "excited/inhibited" in the revised method as following: Page 33-34 in revised manuscript: --"The criteria used to define an "excited" or "inhibited" neuronal response were as follows: firing rates were considered to be significantly different if they were >2 SDs greater or less than baseline averages 66 . Briefly, the average firing rate during each 10-s-duration bin was calculated in 1-min baseline period. Then, the average firing rate of the seizure period or photostimulation was calculated and compared with that of the baseline period to test whether firing rate was >2 SDs greater or less than the baseline average."--

The authors claim that SNr PV cells control TLE seizures via a new "disinhibitory" projection from
SNr PV cells to GABAergic cells in Pf, rather than via the direct projection of SNr to glutamatergic cells. "Quantity analysis showed the number of SNr PV neurons targeting PF glutamatergic neurons was much lower than that targeting PF GABAergic neurons (Fig. S6)." However, SNr PV stimulation inhibits a non-negligible population of Pf GLU cells as well (see Fig 5B). In other words, in light of the results shown in Fig. 5B and Figs. S5 and S6, the overall working model is unclear. The proportion of GABA vs GLU cells in Pf is only 5% in the Pf nucleus (see Fig  S5A); Moreover less than 20% of these GABA Pf cells receive a functional input from SNr (Fig S5, patching results: only 3 GABA cells out of 17 receive SNr inputs). Could this low number be due to sub-optimal recording conditions? What is the rationale for using 4-AP and TTX when recording evoked IPSCs from SNr to Pf thalamus?
Can the authors speculate on how the SNr projection onto just 1% of Pf nucleus is more important for seizure control than the direct targeting of GLU Pf cells by SNr (which would represent a higher number of cells according to Fig 5B, although inconsistent with Fig S6)?
Reply: Thank you very much for all these important comments.
We have also noticed that there seems to be a discrepancy between structural connection and functional connection. Indeed, quantity analysis of viral tracing data showed the number of SNr PV neurons targeting PF glutamatergic neurons was much lower than that targeting PF GABAergic neurons. While, electrophysiological data, including extracellular recording and patch recording, was only used to verify there is a functional input from SNr PV neuron. The success rate of functional connection is extremely low, especially in patch data (only 3 GABA cells out of 17 receive SNr inputs), the possible reason may be: (1) SNr PV neurons may project to a small part of PF GABAergic neurons. In our electrophysiological experiment, it is possible that not all the recorded PF GABAergic neurons actually receive projections from the SNr. (2) The viral-mediated ChR2 expression at the presynaptic part may be low efficient, and the light level is not able to induce GABAergic responses in postsynaptic PF GABAergic neuron 1 . Similar low efficient synaptic transmission in electrophysiological data has been descripted in many previous reports 2-4 .
(3) As reviewer referred, this can also be due to sub-optimal recording conditions. To isolate monosynaptic current, experiments were performed in the presence of TTX and 4-AP to block indirect circuit, similar as previous studies 5, 6 . Together, our anatomical and functional evidence at least indicate that a portion of SNr PV neurons directly innervate the PF GABAergic neurons.
Thus, to make it more clear, we have added some explanation in the revised manuscript as following: Page 15 in revised manuscript: --"The low percentage of PF GABAergic neurons receiving projections from SNr may relate to the following three reasons: (1) SNr PV neurons may project to a small part of the posterior PF GABAergic neurons. (2) The efficacy of virus-mediated ChR2 expression at the presynaptic part is low, and the light level is not able to induce GABAergic responses in postsynaptic PF GABAergic neurons. (3) Sub-optimal recording conditions; to isolate monosynaptic current, experiments were performed in the presence of TTX and 4-AP to block indirect circuit."--In Fig S7,   Reply: Thank you very much for all these important and constructive comments. We totally agree with the reviewer that the chemogenetic activation of PF GABAergic neurons can indirectly inhibit PF glutamatergic neurons, which also blocks pro-epileptic effects of optogenetic activation of SNr-PF GABAergic projections. Indeed, in vivo single-unit recordings showed that photo-activation of SNr-PF PV axons reliably and quickly inhibited the firing rate of 6/12 putative GABAergic neurons recorded in the PF. While the firing change of PF glutamatergic neurons in response to photo-activation of SNr-PF PV axons was heterogeneous: 5/26 neurons were decreased, 6/26 neurons were increased and 15/26 neurons had no change from 6 PV-ChR2 SNr-PF mice (Fig. 5B). Similarly, we also verified this phenennon in KA-induced chronic awake mice (Fig. S6C-E). These data suggested that there is both direct and indirect neural connection between SNr PV neuron and PF glutamatergic neurons. Further, we found that during kindling-induced hippocampal seizures, about 54.5% of GABAergic neurons in the PF decreased their firing rate during seizures (1/21 neuron was excited, 11/21 neurons were inhibited, 9/21 neurons were not changed). Whereas, 50% of glutamatergic neurons in the ipsilateral PF increased their firing rate during seizures (16/33 neurons were excited, 1/33 neuron was inhibited, 15/33 neurons had no response, Fig. S9A-S9C). Optogenetic activation of PF GABAergic neurons inhibited surrounding glutamatergic neurons locally in the Vgat-ChR2 PF mice (Fig. S9D), and direct optogengetic inhibition of PF glutamatergic neurons produced an anti-epileptic effect on seizure development. These data indicated that indirect neural connection between SNr PV neuron and PF glutamatergic neurons may contribute more to the seizure. Although our data cannot support the seizure modulation effect of SNr PV neuron is 100% due to the GABAergic neuron-mediated indirect pathway, it at least demonstrated that PF GABAergic neuron and surrounding PF glutamatergic was involved in seizure modulation. Thus, as reviewer suggested, we have added a working model in the revised Fig. S11 and carefully edited the text to make it clear how SNr PV neurons modulate PF neurons and seizure of TLE.
In addition, we agree with the reviewer that bicuculline is not cell type specific, as it could affect both glutamatergic and GABAergic cells in the PF. Thus, we have indicated this caveat in the revised manuscript.

Revised Fig. S11
Figure S11 ‫|‬ Summary of a disinhibitory nigra-parafascicular neural circuit in seizure in temporal lobe epilepsy.

Page 14 in revised manuscript:
--"The firing change of PF glutamatergic neurons in response to photo-activation of SNr-PF PV axons was heterogeneous: 5/26 neurons were decreased, 6/26 neurons were increased and 15/26 neurons had no change from 6 PV-ChR2 SNr-PF mice (no inter-animal statistical difference, Chi-square test, p=0.5875, Fig. 5B). This data suggested that there might be both direct and indirect neural connection between SNr PV neuron and PF glutamatergic neurons."--

Page 19 in revised manuscript:
--"suggesting PF glutamatergic neurons were a part of the circuit regulating seizure activities of TLE in a bidirectional manner (see summary diagram in Fig. S11)"--Page 20 in revised manuscript: --"suggesting PF glutamatergic neurons were a part of the circuit regulating seizure activities of TLE in a bidirectional manner (see summary diagram in Fig. S11)"--