The Tuberous Sclerosis gene, Tsc1, represses parvalbumin+/fast-spiking properties in somatostatin-lineage cortical interneurons

Medial ganglionic eminence (MGE)-derived somatostatin (SST)+ and parvalbumin (PV)+ cortical interneurons (CINs), have characteristic molecular, anatomical and physiological properties. However, mechanisms regulating their diversity remain poorly understood. Here, we show that conditional loss of the Tuberous Sclerosis (TS) gene, Tsc1, which inhibits mammalian target of rapamycin (MTOR), causes a subset of SST+ CINs, to express PV and adopt fast-spiking (FS) properties, characteristic of PV+ CINs. These changes also occur when only one allele of Tsc1 is deleted, making these findings relevant to individuals with TS. Notably, treatment with rapamycin, which inhibits MTOR, reverses these changes in adult mice. These data reveal novel functions of MTOR signaling in regulating PV expression and FS properties, which may contribute to some neuropsychiatric symptoms observed in TS. Moreover, they suggest that CINs can exhibit properties intermediate between those classically associated with PV+ or SST+ CINs, which may be dynamically regulated by the MTOR signaling.


Introduction
Tuberous Sclerosis (TS) is a disorder that affects multiple organ systems. Roughly half of those diagnosed with TS have autism spectrum disorder (ASD) or intellectual disability (ID), and ~90% exhibit seizures [1][2][3][4] . TS is caused by mutations in the TSC1 and TSC2 genes, which encode the HAMARTIN and TUBERIN proteins, respectively 5,6 . HAMARTIN and TUBERIN proteins dimerize to form the Tuberous Sclerosis (TS) complex which inhibits the activity of the mammalian target of rapamycin (MTOR), a protein complex that is a rheostat for energy homeostasis and is also an important regulator of protein translation 7 . Multiple proteins in the TScomplex/MTOR signaling pathway are either high confidence ASD-causative genes or underlie disorders with high ASD coincidence 8,9 . This has relevance to the high rate of ASD in TS and potentially other TS-Associated Neuropsychiatric Disorders (TANDs), which are common in the syndrome 10 . Uncovering how this pathway regulates neuronal development and function is therefore fundamental to understanding the molecular and cellular underpinnings of ASD and complex neuropsychiatric symptoms in TS.
Accumulating evidence suggests that neuropsychiatric disorders, such as ASD, and associated comorbidities like epilepsy, may be partially caused by changes in cortical GABAergic interneuron (CIN) function and connectivity, which leads to excitation/inhibition (E/I) imbalance in cortical circuits [11][12][13] . While the role of MTOR signaling and Tsc genes on excitatory neurons has been studied for some time, relatively little is known about their roles in CIN development and function [14][15][16] . CINs are the major source of cortical inhibition and are largely derived from the medial and caudal ganglionic eminences (MGE and CGE) 17,18 . PV+ and SST+ CINs are derived from MGE and constitute ~70% of all CINs. These cells have substantially different morphological and physiological properties 19,20 . PV+ CINs exhibit fast-spiking (FS) firing properties and synapse onto soma/axons of excitatory neurons. By contrast, SST+ CINs have regular-spiking (RS) firing properties and target the distal dendrites of excitatory neurons 20,21 .
Progenitor cells in the MGE that express the Lhx6 transcription factor (TF) generate descendants that become either PV or SST CINs [22][23][24][25] . Most studies investigating MGE-derived CIN fate and function have largely focused on the role of TFs 23,24,26,27 , yet little is known about how cellular signaling influences CIN development.
Recent work from us and others highlighted the importance of Pten, another inhibitor of MTOR signaling that acts upstream of Tsc1, on establishing proper numbers of MGE-derived CINs in the cortex 28,29 . Therefore, we speculate that aberrant MTOR signaling in TS might lead to abnormalities in the development and function of MGE-derived CINs.
To investigate this, we conditionally deleted Tsc1 in MGE-derived SST-lineage CINs, which allowed us to assess the impact of Tsc1 loss/MTOR activity starting during early post-mitotic stages. We then investigated the role of Tsc1/MTOR signaling in CIN development, cell fate and physiology. Surprisingly, homozygous loss of Tsc1 caused SST-lineage CINs to aberrantly exhibit cell fate and physiological properties of PV+/FS CINs.
Notably, a subset of SST-lineage CINs resemble PV+/FS CINs in mice both homozygous and hemizygous for Tsc1, suggesting that these observations are relevant to humans with TS. In addition, this phenotype can be rescued by inhibiting MTOR during adult stages, suggesting that drugs currently being studied to treat TS, including rapamycin derivatives, may be effective in treating TS symptoms caused by CIN dysfunction. Overall, our findings demonstrate novel roles for Tsc1 in the development and function of CINs. Notably, we propose that the choice between SST+ and PV+ cell fate and function is mediated in part by non-transcriptional processes, including cellular signaling events, suggesting a new avenue towards understanding these important cell types.

Loss of Tsc1 causes ectopic expression of PV in SST-lineage CINs
To test whether loss of Tsc1 in SST-expressing post-mitotic CINs alters their development, we crossed  with Yate's correction, (WT vs cHet, p = 0.03, cKO vs. WT or cHet, p < 0.0001), n = 3 mice, all groups. Data are represented as mean ± SD in (h) and ± SEM in all other graphs. Scale bars in (f, p) = 100µm. * p < 0.05, **** p < 0.0001. expressed SST was unchanged, there was ~3-fold increase in the % of SST-Cre-lineage CINs that expressed PV in cKO mice (Figs. 1a-f, i, j). These data suggest that SST and PV proteins could be co-expressed in a

SST+ CINs in Tsc1 cKOs exhibit fast-spiking physiological properties
Next, we assessed the physiological properties of SST-lineage CINs in the Tsc1 cKOs. We obtained ex-vivo patch clamp recordings from layer 5 SST-Cre-lineage CINs (tdTomato+) in WT, cHet and cKO mice (example cells, Fig. 2a, and recordings, Fig. 2b). WT CINs had RS physiological properties, including strong spike frequency accommodation (SFA), wider spikes, relatively slow action potential (AP) rate of rise (max dV/dt) and relatively small fast afterhyperpolarization (fAHP) amplitudes (Fig. 2), as expected 19,20,34 . Consistent with the increased soma size, known to be regulated by Tsc1/MTOR signaling 34,35 , and the increase in cKO soma   output might result in increased spontaneous inhibitory transmission. This reduction in spontaneous inhibitory output could be caused by a reduction in numbers of inhibitory synapses and/or probability of release at these synapses. Towards elucidating the respective contributions of these various possibilities, we added TTX (1µM) to the recording solution, to remove large amplitude, AP-dependent IPSCs (and therefore any effects of changes in excitability). Similar to sIPSCs, the frequency and amplitude of miniature IPSCs (mIPSCs) were lower in Tsc1 cKOs (Figs. 4c, d). These findings suggested that the loss of Tsc1 reduced the strength of inhibitory synaptic transmission by altering the number and/or release probability at inhibitory synapses.
While we found significant reductions in spontaneous inhibitory output to cortical pyramidal neurons, we could not estimate the specific contribution of SST+ CINs to this reduction by analyzing sIPSCs and mIPSCs. To  The axons of PV+/FS CINs target the soma and axon initial segment of pyramidal neurons, while the axons of SST+ CINs target their distal dendrites 20,21 . Thus, we tested whether Tsc1 deletion caused aberrant axon targeting (proximal dendrite/soma vs. distal dendrite) in a subset of SST-lineage CINs. For this, we obtained paired patch clamp recordings from connected layer 5 pyramidal neurons (Pyr) and tdTomato+ SST-Crelineage CINs (Fig. 5a) and analyzed the unitary IPSCs (uIPSCs) in postsynaptic Pyr neurons evoked by action potentials in presynaptic SST+ CINs (Fig. 5b). Since uIPSCs generated in response to SST+ CINs arrive at the distal dendrites, the amplitude and kinetics of these currents are significantly different from IPSCs originating near the soma of pyramidal neurons 39

Cell autonomous role for Tsc1 in regulating PV expression in CINs
Since Tsc1 cKOs exhibited reduced inhibitory tone in the neocortex (Fig. 4), we asked if the increased PV expression in SST-Cre-lineages is cell autonomous, as PV expression can be modulated by circuit activity 41,42 .
We utilized an MGE transplantation assay that introduces MGE progenitors in small numbers to a WT cortex for in vivo maturation 43 . MGE tissue from WT, Tsc1 Floxed/+ or Tsc1 Floxed/Floxed was isolated from E13.5 embryos; dissociated cells were transduced with DlxI12b-Cre expressing lentiviruses (Fig. 6a). The DlxI12b enhancer biases expression to GABAergic neurons and has been used to express genes efficiently in developing and mature CINs 22,28,44 . These MGE cells were then transplanted into WT neonatal cortex and allowed to develop.  (Figs. 6b-d, f, g). Finally, we expressed the human TSC1 gene from the same lentiviral vector, to test if human TSC1 could compliment the mouse Tsc1 depleted CINs and found that it was able to rescue the increase in soma size and PV expression (Figs. 6e-g).

Cre-lineage CINs
Based on the preceding results, we hypothesized that Tsc1 normally represses PV-like/FS properties in SST+ CINs by inhibiting MTOR activity. We thus administered the MTOR inhibitor, rapamycin, to test whether this treatment might reverse/rescue the effects of Tsc1 loss in young adult (eight-week old) mice. Tsc1, cHets and cKOs were treated with rapamycin or vehicle starting at eight weeks of age, for five consecutive days (Fig. 7a).
A day after the last dose, mice were assessed for soma size, PV expression and electrophysiology. To assess rapamycin's efficacy, we probed for pS6 in tdTomato+ CINs. Vehicle-treated Tsc1 cKO CINs had increased coexpression of pS6 compared to the cHet groups and rapamycin treatment significantly reduced these levels ( Supplementary Figs. 7a-e). Notably, rapamycin treatment did not alter the cell density of tdTomato+ CINs or those expressing SST (Supplementary Figs. 7f, g).
As expected, vehicle treated cKO CINs exhibited increased soma size and PV expression compared to the cHet groups (Figs. 7b-d). Rapamycin treatment significantly reduced soma size of SST-lineage CINs in cKOs but not in the cHets (Figs. 7b, c). We also observed reduced PV expression in rapamycin-treated cKOs (Figs. 7b, d). However, rapamycin treatment did not alter PV expression in SST-Cre-lineage cHets (Figs. 7b, d).  the fraction of FS SST-Cre-lineage FS CINs in WT mice (Fig. 2h, inset). Rapamycin-treatment also failed to reverse some effects of Tsc1 loss, including altered AP properties in cKOs (Supplementary Fig. 8). Overall, the five-day rapamycin treatment in adult cKO mice decreased the aberrant PV expression and partially reversed shifts towards FS phenotypes in SST-Cre-lineage CINs.

Discussion
Imbalances in neuronal excitation/inhibition (E/I) are implicated in neuropsychiatric disorders 11,12,45,46  Previous studies in a mouse model of TS emphasized that dysregulation of MTOR signaling primarily affects excitatory circuits 14 , and alters E/I balance through effects on excitatory synapses. While one report examined CINs with regards to TS 51 , little was known about changes in inhibitory circuits. In this context, we assessed cellular and physiological properties of CINs and inhibitory synapses in Tsc1 cHets and cKOs. We discovered that loss of Tsc1 in SST-lineages resulted in aberrant expression of PV and the voltage-gated potassium channel, Kv3.1, in a subset of SST-lineage CINs. This, in turn, shifts their physiological properties from RS to a continuum encompassing both RS and FS properties. Notably, not all SST-lineage CINs adopted these properties, showing that some are less susceptible. This is particularly interesting as recent data has shed new light on subpopulations of SST+ CINs, with distinct molecular identities and functions [52][53][54] . Future studies should aim at determining whether the cohort of SST-lineage CINs in which MTOR signaling normally represses PV and Kv3.1 expression as well as FS properties, represents a subtype with specific molecular and/or functional properties. Tsc1 cHets exhibited similar (but less severe) phenotypes, suggesting that these findings are relevant to TS. Moreover, TSC1/2 homozygous mutants have been identified in neural progenitors via a combination of germline and somatic mutations in a TS patient 55 , suggesting the more severe phenotypes we observed in Tsc1 cKO mice may also be relevant to some cases of TS.
This study suggests new ideas about how some aspects of CIN cell fate and function are regulated, particularly in post-mitotic/maturing CINs. Previous studies identified TFs that are expressed in SST+ CINs but not in PV+ CINs, as well as TFs that can alter the ratios of SST+ and PV+ CINs [22][23][24]27,[56][57][58] . Of note, we are not aware of a TF that is uniquely expressed in PV+ but not SST+ CINs 59 . One hypothesis to explain this is that PV+ CIN identity is a default state of MGE-derived CINs, and this fate is repressed by TFs restricted to SST-lineages 59 .
Furthermore, perhaps during the maturation of PV-lineage CINs, MTOR activity is induced to promote PV expression and FS properties, i.e. when PV expression first becomes evident during adolescence. This may be achieved as a consequence of activating certain growth factor receptors or potentially via synaptic activity 7,60 in prospective PV-lineages. However, given that Tsc1/2 and other components of the MTOR pathway are expressed in most cell types, there would need to be unique factors coupling MTOR signaling in PV-lineages that are different in SST-lineages. These ideas will be investigated in future studies.
Most PV+ CINs have FS physiological properties. In contrast, most SST+ CINs exhibit a RS physiology, although a class of "quasi-fast-spiking" SST+ neurons was recently described 52 . Moreover, these two neuronal groups are believed to play distinct and dissociable roles in circuit function [61][62][63][64][65] . Our data suggest new possibilities about these classes. First, CINs exhibit intermediate properties after Tsc1 loss, which fall along a continuum between the classic SST+ and PV+ distinct groups of CINs. Second, the specification of these classes may be more dynamic in later developmental periods than previously thought. In particular, rapamycin was able to reverse the expression of PV and FS properties in mature SST-lineage cKO CINs, suggesting that some aspects of CIN cell fate phenotypes are malleable by signaling-dependent plasticity mechanisms that remain to be elucidated.
The extent to which changes in molecular (expression of PV vs. SST) and physiological (FS vs.RS) properties correlate with changes in cortical function remains a major question. Differential expression of ion-channels, like delayed-rectifying Kv3 channels, underlie the fast-spiking properties of PV+ CINs 37,66 . Loss of Tsc1 leads to an increase in the occurrence of "dual identity" CINs, which express both PV and SST. Furthermore, SSTlineage CINs lacking Tsc1 have increased expression of Kv3.1 and the properties of these CINs are shifted towards a FS physiology. Interestingly, deletion of FMRP, another monogenic ASD-risk gene, inhibits protein translation downstream of MTOR, of which the Kv3.1 channel is known to be a target in auditory brain stem neurons 47. These converging findings suggest that modulators of the Kv3.1 channel could be potential targets for treating cellular and circuit abnormalities in TS. Future work will be necessary to understand whether Tsc1 deletion induced shifts in CIN cell fate and physiology underlie behavioral abnormalities in mutant mice and humans diagnosed with TS, which may lead to new therapeutic targets.
In summary, we provided evidence that Tsc1-inhibition of MTOR represses PV+/FS properties in a cohort of SST-Cre-lineage CINs. Notably, it suggests that regulation of some CIN molecular and physiological properties can be initiated by non-transcriptional events, i.e. MTOR signaling . However, whether transcription is targeted downstream of MTOR signaling is not yet known. Moreover, we identified specific alterations in GABAergic CINs and underlying molecular mediators (e.g. MTOR and Kv3.1) that could plausibly contribute to neurocognitive symptoms of TS. Future studies should assess whether similar abnormalities occur in humans diagnosed with TS (e.g. by immuno-histochemistry in post-mortem tissue) and/or can be reversed by rapamycin analogs (e.g. in neurons derived from human induced pluripotent stem cells), to more directly implicate these findings about CINs in the pathogenesis and/or treatment of TS.

Animals
All mouse strains have been published: Ai14 Cre-reporter 68

Cell Counting
To determine cell density (cells/mm 2 ), we counted the number of cells in a given section and then divided by the area of that region. To calculate the % of tdTomato + cells that co-labeled with specific markers, we divided the number of co-labeled cells by the total number of tdTomato + cells. For cell transplants, all tdTomato + cells were counted in the neocortex from all sections in a rostral to caudal series. For cell fate counts, only transplants where at least 50 tdTomato + cells could be counted were used for analysis. For soma size quantification, the perimeter of each tdTomato + cell's soma was traced in Image J, and at least 25 cells were measured and averaged for each mouse.

Acute cortical slice preparation
Adult mice of either sex (P45-P60 days) were anesthetized with an intraperitonial injection of euthasol and transcardially perfused with an ice-cold cutting solution containing (in mM) 210 sucrose, 2.  Electrophysiology data were recorded using Multiclamp 700B amplifier (Molecular Devices). Voltages have not been corrected for measured liquid junction potential (~8 mV). Upon successful transition to the whole-cell configuration, the neuron was given at least 5 min to stabilize before data were collected. Series resistance and pipette capacitance were appropriately compensated, before each recording. Series resistance was usually 10-20 MΩ, and experiments were terminated if series resistances exceeded 25 MΩ.

Electrophysiology protocols and data analysis
All data analyses were performed using custom routines written in IGOR Pro (Wavemetrics). Code is available upon request. Resting membrane potential (RMP) was measured as the membrane voltage measured in current clamp mode immediately after reaching the whole-cell configuration. Input resistance (Rin) was calculated as the slope of the linear fit of the voltage-current plot generated from a family of hyperpolarizing and depolarizing current injections (-50 to +20 pA, steps of 10 pA). Firing output was calculated as the number of action potentials (APs) fired in response to 800 ms long depolarizing current injections (25-500 pA). Firing frequency was calculating as the number of APs fired per second. Rheobase was measured as the minimum current injection that elicited spiking. Firing traces in response to 50 pA current above the rheobase were used for analysis of single AP properties-AP threshold, maximum dV/dt (rate of rise of AP), AP amplitude, AP halfwidth and fast after hyperpolarization (fAHP) amplitude. Threshold was defined as the voltage at which the value of third derivative of voltage with time is maximum. Action potential amplitude was measured from threshold to peak, with the half-width measured at half this distance. Fast after hyperpolarization was measured from the threshold to the negative voltage peak after the AP. Index of spike-frequency accommodation (SFA) was calculated as the ratio of the last inter-spike interval to the first inter-spike interval.
Coefficient of variance (CV) for inter-spike interval (ISI), AP amplitude and AP half-width was calculated as the ratio of standard deviation to the mean. Recorded CINs in all genotypes were classified as fast-spiking or regular-spiking based on electrophysiological properties. Specifically, CINs were classified as fast-spiking if the AP half-width was < 0.5 ms, firing frequency > 50 Hz, fAHP amplitude was > 14 mV and SFA was < 2. The correspondence between PV expression and fast-spiking electrophysiological properties was confirmed in a subset of CINs by quantification of PV staining in biocytin-filled neurons (16 of 21 biocytin-labelled CINs classified as fast-spiking were PV+).
Spontaneous inhibitory currents were recorded for 5 min with neurons voltage clamped at +10 mV. Miniature inhibitory currents were recorded in the presence of TTX. Spontaneous and miniature inhibitory currents were analyzed off-line using Clampfit (pClamp, Molecular devices) event detection.
For Cre-dependent expression, 600 nl of AAV-DIO-ChR2-eFYP virus was bilaterally injected into P30-40 SST-Cre mice. We waited at least 4 weeks after virus injection before preparing brain slices. To measure optogenetically evoked inhibitory currents (oIPSCs), we voltage-clamped Layer 5 pyramidal neurons at +10 mV. We stimulated ChR2 using 5 ms long single or multiple light pulses (maximum light power, 4 mW/mm 2 ) generated by a Lambda DG-4 high-speed optical switch with a 300 W Xenon lamp (Sutter Instruments) and an excitation filter centered around 470 nm, delivered to the slice through a 40X objective (Olympus). Frequency dependent attenuation (20 Hz) and paired-pulse ratios of oIPSCs were measured at the maximum light power.
To measure unitary inhibitory currents (uIPSCs), dual whole cell recordings were obtained from connected layer 5 tdTomato expressing (SST-Cre lineage or PV-Cre lineage) CINs and adjacent layer 5 pyramidal neurons. We obtained current clamp recordings from tdTomato+ CINs and voltage clamp recordings (+10 mV) from pyramidal neurons. Short duration (1-2 ms) depolarizing current steps (1-1.5 nA) were injected into the CINs to elicit single action potentials and uIPSCs were recorded in postsynaptic pyramidal neurons. Average of 20-30 uIPSCs were used to calculate the peak amplitude (from the baseline to the peak of uIPSC), rising slope (slope of a fitting line from 10-90% after onset of uIPSC) and total charge of uIPSCs (integral of uIPSCs from the onset until the trace returned completely to the baseline).

Computer simulations
Simulations were performed using the NEURON

MGE cell transplantation
Lentiviral transduction of MGE cells followed by intracranial transplantation was performed as previously described 43 . First, HEK293T cells were transfected using Lipofectamine2000 (Thermo Fischer Scientific) with four plasmids to generate lentivirus particles. E13.5 MGEs from individual embryos were dissected in ice-cold HBSS and then kept on ice in DMEM media (containing 10% fetal bovine serum). MGEs were then mechanically dissociated with a p1000 pipette tip and transduced with lentivirus. To perform transductions, dissociated MGE cells were mixed with pre-warmed media that was at physiological pH and was incubated with ~10-20 μls of concentrated lentivirus, with titers ranging from 1X10^8 -1X10^10 infectious units/ml. They were next incubated at 37°C for 30 minutes, with intervals of agitation. Since the MGE cells had a Credependent reporter, Ai14 68 , only MGE cells transduced with Cre-expressing lentiviruses were visible in the host neocortex after transplantation. This combinatorial method allows the transduced/transplanted cells to express a strong reporter, i.e. tdTomato expressed from the beta-actin promoter, while maintaining cell type specificity. Cells were then pelleted in a tabletop centrifuge at low speed (700xg, ~4 minutes) and washed 3 times with media followed by trituration to disperse cells between each wash to remove excess virus. The final pellet was left under a few μls of media, put on ice, and the remaining media was removed with a fine point kim wipe before the injection needle was loaded. For injections, a glass micropipette with a 45º beveled tip, ~50 μm outer diameter, was preloaded with sterile mineral oil and cells were then front-loaded into the tip of the needle using a plunger connected to a hydraulic drive (Narishige MO-10) that was mounted to a stereotaxic frame.
Pups were anesthetized on ice for 1-2 minutes before being placed on a mold for injections. Each pup received 3-6 injections of cells, at 70 nl per site, in their right hemisphere. These sites were about 1 mm apart from rostral to caudal and medial to lateral and were then injected into layers V-VI of a P1 WT neocortex. After injections, pups were allowed to recover and then put back with their mother. The transplanted mice were sacrificed at 35 days post-transplant and transcardially perfused with PBS followed by 4% PFA. Brains were then post-fixed in 4% PFA for a short interval, ~30 minutes, and then sunk in 30% sucrose before embedding in OCT.

DNA vector generation
The DlxI12b-BG-IRES-Cre lentiviral vector was previously described . To generate the DlxI12b-BG-hTSC1-IRES-Cre lentiviral vector, the human TSC1 gene was PCR amplified from a vector containing the hTSC1 cDNA (Addgene) 70 . The primers (Forward 5'-GAGAGCTAGCATGGCCCAACAAGCAAATG-3' and Reverse 5'-GAGATGTACATTAGCTGTGTTCATGATG-3'), were used to introduce NheI and BsrGI restriction enzyme sites, respectively (underlined). Next the PCR product and the DlxI12b-BG-IRES-Cre vector were digested with the appropriate enzymes and ligated. The vector was verified by restriction digest and sequencing.

Rapamycin treatment
Rapamycin (LC Laboratories, Woburn, MA, USA) was dissolved in ethanol at a concentration of 20 mg/ml and stored at -20°C. On the day of the injections, drug was resuspended in saline (containing 0.25% PEG and 0.25% Tween-80). Mice received rapamycin solution (10 mg kg −1 ) or an equal volume of vehicle by intraperitoneal injection once daily for five consecutive days.