SUMMARY
The circuitry of the striatum is characterized by two organizational plans: the division into striosome and matrix compartments, thought to mediate evaluation and action, and the direct and indirect pathways, thought to promote or suppress behavior. The developmental origins of and relationships between these organizations are unknown, leaving a conceptual gap in understanding the cortico-basal ganglia system. Through genetic fate mapping, we demonstrate that striosome-matrix compartmentalization arises from a lineage program embedded in lateral ganglionic eminence radial glial progenitors mediating neurogenesis through two distinct types of intermediate progenitors (IPs). The early phase of this program produces striosomal spiny projection neurons (SPNs) through fate-restricted apical IPs (aIPSs) with limited capacity; the late phase produces matrix SPNs through fate-restricted basal IPs (bIPMs) with expanded capacity. Remarkably, direct and indirect pathway SPNs arise within both aIPS and bIPM pools, suggesting that striosome-matrix architecture is the fundamental organizational plan of basal ganglia circuitry organization.
INTRODUCTION
The striatum, the gateway and largest nucleus of the basal ganglia, receives massive inputs from all areas of the cerebral cortex and gives rise to the largest output circuits of the basal ganglia by way of multi-synaptic pathways leading to the brainstem and thalamus. Neural activity in the striatum itself is influenced by dopamine-containing afferents from the substantia nigra, serotonergic afferents from raphe nuclei, cholinergic inputs from the brainstem, and other neuromodulatory circuits. The striatum thus constitutes a key neural substrate through which diverse functionally specialized regions of the neocortex converge to modulate not only downstream motor programs for the initiation of voluntary behaviors, but also neural systems underpinning emotion, motivation, evaluation and learning (Graybiel, 2008; Grillner and Robertson, 2015; Hikosaka et al., 2014).
The majority of neurons in the striatum are output spiny projection neurons (SPNs). These are known to follow two fundamental organizational schemes. First, the striatum has a striking compartmental organization by which labyrinthine neurochemically specialized zones called striosomes (‘striatal bodies’) lie embedded within the much larger matrix compartment (Graybiel and Ragsdale, 1978). This striosome-matrix architecture contributes to the functional input and output connectivity of the striatum (Crittenden and Graybiel, 2011; Grillner and Robertson, 2015). The SPNs of striosomes receive preferential input from particular limbic regions and have privileged output to the dopamine-containing neurons of the substantia nigra and indirectly to the lateral habenula, thus engaging crucial dopamine- and serotonin-related neuromodulatory systems (Fujiyama et al., 2011; Jiménez-Castellanos and Graybiel, 1989; Watabe-Uchida et al., 2012; Crittenden et al., 2016; Eblen and Graybiel, 1995; Brimblecombe and Cragg, 2017). By contrast, SPNs of the much larger matrix compartment receive inputs from sensorimotor and associative cortical regions and in turn project to the main pallidonigral output nuclei of the basal ganglia that target downstream premotor regions and thalamocortical circuits (Donoghue and Herkenham, 1986; Flaherty and Graybiel, 1994; Graybiel, 2008; Hikosaka et al., 2014). Functionally, striosomal circuits have been implicated in evaluative functions related to affective control, cost-benefit decision-making, and reinforcement-based behaviors. By contrast, the matrix appears to be involved in the translation of cortical action plans and strategies to basal ganglia circuits involved in action execution (Amemori et al., 2011; Cui et al., 2014; Friedman et al., 2015; Fujiyama et al., 2011; Gerfen, 1992; Lopez-Huerta et al., 2015; White and Hiroi, 1998; Parthasarathy et al., 1992; Giménez-Amaya and Graybiel, 1991; Graybiel et al., 1994). The developmental basis of this striosome and matrix segregation is not well understood.
Superimposed upon striosome-matrix compartmentalization is a split between the projections of SPNs to the main basal ganglia output nuclei, the pallidum and the substantial nigra pars reticulata (SNr) (Gerfen, 1992; Kim and Hikosaka, 2015). SPNs projecting to the internal pallidum and SNr constitute the so-called direct pathway SPNs (dSPNs), conventionally considered to release basal ganglia-mediated inhibition or inhibition-excitation sequences and to promote action. By contrast, the SPNs projecting to the external pallidum give rise to the indirect pathway (iSPNs), considered to inhibit actions, potentially those with unwanted characteristics (Freeze et al., 2013; Albin et al., 1995). The direct-indirect pathway dichotomy has been elegantly formulated as a center-surround system in which wanted movements are promoted by the direct pathway and competing, interfering motor programs are inhibited (Mink, 1996)). By virtue of its large size and output organization, the matrix compartment is the main source of the direct and indirect pathways, but striosomes as well as matrix contain both dSPNs and iSPNs, at least as categorized by the expression of dopamine D1 (in dSPNs) and D2 (in iSPNs) receptors (Crittenden and Graybiel, 2011; Fujiyama et al., 2011; Brimblecombe and Cragg, 2017; Salinas et al., 2016).
Despite strong evidence for the coexistence of the striosome and matrix compartments and the direct and indirect pathways, remarkably little is known about the relationship between these two fundamental axes of striatal organization. A clue to the potential importance of developmental mechanisms underlying the compartmental divisions first came from studies demonstrating that SPNs making up striosomes and matrix have different birthdates, with striosomal SPNs being generated earlier than those within the matrix (Newman et al., 2015; van der Kooy and Fishell, 1987). Here we sought to link this evidence to neural progenitors and lineage mechanisms known to be crucial in striatal developmental cascades (Rubenstein and Campbell, 2013). The neuroepithelium of the lateral ganglionic eminence (LGE) of the embryonic ventral telencephalon consists of diverse types of progenitors that emerge and differentiate across different stages of striatal neurogenesis (Turrero García and Harwell, 2017). These include regenerative radial glial progenitors (RGs), which give rise to apical intermediate progenitors (aIPs) such as short neural progenitors (SNPs) and subapical progenitors (SAPs), and basal intermediate progenitors (bIPs) (Pilz et al., 2013). Although certain lineage relationships and neurogenic capacity of some of these progenitor types have been characterized by live imaging at mid-gestation (e.g., Pilz et al., 2013), their developmental trajectory through the course of LGE neurogenesis and especially their link to different SPN types and striatal circuit organization remain unknown. Tapping into the molecular mechanisms and transcription factors expressed in different progenitor types, we performed systematic genetic fate mapping to resolve different progenitor types and trace their developmental trajectories from early lineage progression to the distinct SPN types that they give rise to.
Here, we demonstrate that striosomal and matrix SPNs are sequentially generated from a RG lineage program through sequential allocation of distinct types of IPs with different amplification capacities. Whereas the early phase of this program produces striosomal SPNs through fate-restricted Ascl1+ aIPSs with limited neurogenic capacity, the late phase amplifies matrix SPN production through the additional deployment and amplification of fate-restricted Ascl1+/Dlx1+ bIPMs with expanded neurogenic capacity. During the final phase, Dlx1/2+ bIPs specifically generate projection neurons of the annular compartment adjoining striosomes (A cells). Remarkably, a similar temporal and progenitor type split of striosomal and matrix SPN production was not observed for the genesis of D1- and D2-receptor bearing direct and indirect pathway SPNs. We found that each of these pathways is derived from both aIPSs and bIPMs throughout the course of RG lineage progression. These findings suggest that a primary lineage program within the LGE gives rise to striosome-matrix compartmentalization, and superimposed upon this organization, a secondary and different mechanism gives rise to direct and indirect pathway SPNs within each compartment. Thus the major components of striatal architecture are rooted in a radial glia lineage program differentiating future striosome and matrix compartments at the inception of striatal development, and they originate from sequential phases of lineage progression and from distinct intermediate progenitor types. These findings establish a novel framework for exploring the assembly of cortico-basal ganglia circuitry and should provide new clues to the differential vulnerabilities of the striosome-matrix and direct-indirect pathways in human disease states.
RESULTS
Early LGE Radial Glial Cells Sequentially Give Rise to Both Apical and Basal Intermediate Progenitors and to Both Striosomal and Matrix SPNs
The initial formation of LGE around embryonic day (E) 9.5 along the subpallium is followed shortly by the onset of neurogenesis, when proliferative neuroepithelial cells (NEs) begin to transform into neurogenic radial glial cells (RGs) (Sousa and Fishell, 2010). Although all striatal SPNs are thought ultimately to be generated from LGE RGs, recent studies indicate the existence of multiple types of progenitors with distinct molecular and morphological characteristics that are derived from RGs at different embryonic times (Pilz et al., 2013; Turrero García and Harwell, 2017). These include aIPs such as SNPs and SAPs), and also bIPs. How these different progenitor types relate to distinct striatal SPN types remains largely unknown. We used several inducible CreER drivers to carry out comprehensive genetic fate mapping of multiple LGE progenitor types.
Among the markers expressed in telencephalic progenitors (i.e., pallium and subpallium), the anti-proliferative protein TIS21 is unique in its specific expression in all progenitors undergoing neurogenic divisions, including subsets of RGs, aIPs, and bIPs (Attardo et al., 2008). To fate map different types of neurogenic progenitors, especially self-renewing RGs in the process of neurogenesis, we generated a knock-in Tis21-CreER driver and demonstrated that TM-induced activation of the Ai14 reporter reliably and efficiently captured neuron-producing progenitors (Figures 1 and S1). The Tis21-CreER driver thus provided a unique and powerful fate-mapping tool for LGE neurogenic RGs starting from the very onset of neurogenesis. This enabled tracing of the RG lineage progression through sets of embryonic IPs to their SPN progenies in the postnatal striatum (Figures 1A and 1B). At the early onset of neurogenesis ~E10, 8-hour pulse-chase of Tis21+ progenitors labeled both RGs and aIPs (Figures 1C, 1F and 1G). We could not strictly distinguish between SNPs and SAPs, as these aIP subtypes are defined by their cell division patterns best captured by live imaging experiments (Pilz et al., 2013). Importantly, 48 hours after E10 TM induction (i.e., at E12), RFP+ RGs and aIPs remained abundant in the ventricular zone (VZ), indicating that RGs self-renewed and continued to generate aIPs. Sparse labeling allowed identification of elongated cell clusters, which likely represented clonal descendants from an initial neurogenic RG (Figure 1D, H, I; Figure S1D). Remarkably, even 96 hours after E10 TM induction (i.e., at E14), RFP+ RGs still remained in VZ, indicating persistent self-renewal (Figures 1E, 1J-1L, S1E, and S1F). Particularly large cell clones were frequent, which consisted of a “founder RG”, multiple Ascl1+ IPs, and larger number of postmitotic neurons likely engaged in migration along the RG radial fiber (Figures 1L and S1F, Supplemental Movie 1). Whereas E10-labeled RGs gave rise to aIPs from E10-E12 (Figures 1D, 1H and 1I), they made a transition to generating mainly bIPs in the subventricular zone (SVZ) by E14 (Figures 1E and 1J-1L).
To determine the identity of SPNs derived from early Tis21+ RGs, we assayed the striatum on postnatal day (P) 28 (Figures 1M–1O). Both striosomal (S) and matrix (M) SPNs were generated, roughly in a 1:4 ratio (Figure 1O). To further probe the lineage relationship between S and M cells, we combined genetic fate mapping of RGs at E10 and consecutive birth dating by EdU at E11 and BrdU at E14. We found that RFP+ S cells were predominantly double-labeled by EdU at E11, whereas RFP+ M cells were predominantly double-labeled by BrdU at E14 (Figures 1P–1W). Together, these results suggest that along the lineage progression, the same early Tis21+ RG population first generates S cells and then M cells, likely via producing several types of IPs.
A Set of Early Apical Intermediate Progenitors Are Fate-Restricted to Generate Striosomal SPNs
The proneural protein ASCL1 is restricted to more differentiated neurogenic progenitors rather than self-renewing RGs and is important in coordinating the balance between neural progenitor proliferation and cell cycle exit toward neurogenesis (Casarosa et al., 1999; Guillemot and Hassan, 2017). ASCL1 expression exhibits two distinct dynamic patterns: an oscillatory pattern of relatively low ASCL1 levels correlating with cell-cycle progression and proliferation, versus a sustained expression at higher levels driving neurogenic cell division and neuronal differentiation (Imayoshi et al., 2013; Imayoshi and Kageyama, 2014). Thus the Ascl1-CreER driver is an effective tool for selectively labeling progenitors biased towards adopting a neurogenic path (Sudarov et al., 2011; Figures 2A–2D). Indeed, throughout E10 LGE, we detected both low and high levels of ASCL1 immunoreactivity in cells within the VZ, suggesting intermingling of cells in various proliferative (oscillatory expression) and neurogenic (sustained expression) states (Figures 2G, 2H, S2A and S2B). We used relatively low-dose TM induction to achieve sparse labeling and to target progenitors with high-level Ascl1 expression, which we verified by co-labeling with anti-ASCL1 antibody (Figure 2H). A brief 8-hour pulse-chase labeling with TM induction in E10.5 Ascl1-CreER;Ai14 embryos reliably labeled aIPs with end-feet at the ventricle surface, with or without basal radial fibers (Figure 2D). Importantly, 18 and 24 hours following induction at E10.5, few or no RFP+ progenitors remained in the VZ, in sharp contrast to Tis21+ RGs (Figure 1D), indicating that E10 Ascl1+ progenitors are aIPs, as opposed to RGs endowed with extensive capacity for self-renewal. Ascl1+ aIPs did not express multi-potency markers such as Sox2 (Figure S2C) or the postmitotic neuron marker BIII-tubulin 1(TuJ1; Figures S2C and S2D). It is possible that a lower level of Ascl1 might have been expressed in RGs as well as in aIPs, which may not have been captured by low-dose TM induction. Newly postmitotic RFP+ cells expressed typical SPN markers such as CTIP2 as they reached the striatum, but not Nkx2.1, a marker of GABAergic interneurons derived from the medial ganglionic eminence (MGE), suggesting that all fate-mapped cells were SPNs (Figures S2ES2H).
We then examined the types of mature SPNs in P28 striatum derived from E10.5-labeled Ascl1+ aIPs (Figures 2J–2N). Strikingly, early Ascl1+ aIPs gave rise almost exclusively to SPNs located within striosomes. These neurons exhibited spiny dendrites and clustered into relatively discrete, widely distributed zones that were positive for the striosome marker mu opioid receptor 1 (MOR1) and negative for the matrix marker CalDAG-GEFI (CDGI) (Figures 2J–2L; Supplemental Movies 2 and 3) (Kawasaki et al., 1998). Both confocal microscopy (Figures 4A and 4C) and serial two-photon tomography (STP; Figure S7; Supplemental Movies 2-4) of the labeled brains demonstrated their morphology and showed that these early aIP-derived SPNs (herein designated as S cells) projected their axons to the dopamine-containing domain of the substantia nigra pars compacta (SNc), a defining feature of striosomal SPNs (Gerfen, 1984; Jiménez-Castellanos and Graybiel, 1989; Crittenden et al 2016). The small fraction of SPNs located in matrix might correspond to the cells described as “exo-patch SPNs” (Smith et al., 2016), long seen in most strisomal preparations (Graybiel and Hickey, 1982; Newman et al., 2015; Graybiel, 1984), likely cells that have not fully completed migration to striosomes but are S cell in type, as the axons of labeled SPNs formed dense projections to SNc, characteristic of S cells. Indeed, throughout the period from E9.5 to E12.5, these Ascl1+ aIPs almost exclusively generated S cells and not M cells (Figures 2P, 2Q, 5 and S4B).
To further verify that S cells were indeed born from Ascl1+ aIPs within this time-window, we combined TM induction at E10.5 followed with BrdU birth dating of their neuronal progeny at several time points (in different litters), including at 4, 24, and 48 hours after TM induction (Figure 2R). These experiments demonstrated that the peak of S cell production from E10.5 aIPs was between E10.5 and E11.5, followed by a sharp decline before E13.5 (Figure 2R). Together, these findings suggest that the early cohorts of Ascl1+ aIPs were fate-restricted to produce striosomal SPNs from E10 to ~E12, and confirmed that early Ascl1+ aIPs do not linger in the VZ for >24 hours.
As the homeobox gene Dlx1/2 is also implicated in LGE neurogenesis and striatal development (Anderson et al., 1997), we performed similar fate mapping using the Dlx1-CreER driver (Taniguchi et al., 2011). Surprisingly, we found that during the early phase of LGE neurogenesis (E10.5), Dlx1 was not expressed in progenitors in VZ but almost exclusively confined to postmitotic neurons in the mantle zone (MZ; Figures 2E and 2F). Fate mapping of these early-born (E9.5-E12.5) Dlx1+ neurons to the mature brain showed that they were almost exclusively S cells, similar to those deriving from Ascl1+ aIPs during the same embryonic time (also see Figure 5). This result suggests that Dlx1 likely acts down-stream and/or subsequent to Ascl1 during this early phase, possibly in regulating the migration and maturation of postmitotic SPNs.
A Set of Basal Intermediate Progenitors Are Fate-Restricted to Generate Matrix SPNs
The nearly exclusive fate restriction of early aIPs for S cell production prompted us to examine later phases of LGE neurogenesis from different types of progenitors. At E14.5, 8 hour pulse-chase in Tis21-CreER;Ai14 embryos continued to label VZ RGs with apical end-feet (Figures 3A and 3E). In addition, large numbers of Ascl1+ bIPs were labeled in the SVZ (Figures 3A and 3F), which were frequently engaged in symmetric cytokinesis (Figure 3G), suggestive of neurogenic cell division. Among the total RFP labeled cells, ~5% were putative RGs (located in VZ with end-feet and expression of the proliferation marker Sox2), ~70% were bIPs (Ascl1+ in SVZ with no end-feet); and the remaining ~25% of Ascl1− cells in SVZ likely represented other types of bIPs (e.g., Dlx1+) as well as new-born postmitotic neurons (Figures 3H and 3I). Notably, in addition to RGs, 24-hour and 72-hour pulse-chase showed the generation of almost exclusively bIPs and not of aIPs (Figures 3B–3D and S3A). In particular, 72-hour pulse-chase demonstrated that Tis21+ RGs continued multiple rounds of self-renewal during the late stage of embryonic development with ongoing generation of Ascl1+ progenitors in SVZ (Figures 3C, 3D and S3B). These results suggest that during the late phase of LGE neurogenesis (from E14.5 on), Tis21+ RGs persist to the end of neurogenesis (Figures S3E and S3F) and begin to predominantly generate a new population of bIPs, possibly in addition to production of a much smaller number of aIPs (see below).
For a more detailed analysis of the neuronal output of Ascl1+ bIPs, we further carried out analogous embryonic fate mapping in E14.5 Ascl1-CreER;Ai14 mice. Eight-hour pulse-chase of Ascl1+ IPs labeled few aIPs and many bIPs, which were identified as progenitors located in the SVZ without apical end-feet (Figure 3K). Importantly, some E14.5-labeled bIPs remained in the SVZ at 24 hours post-induction (Figures 3L S3C and S3D). As the cell-cycle length of LGE IPs at this time is only ~12 hours, this finding corroborated results from live imaging experiments (Pilz et al., 2013) and suggested that some bIPs may serve as transit-amplifying progenitors that proliferate in the SVZ before neurogenesis (Figure 3J). To identify the SPN types derived from late-phase Ascl1+ bIPs, we analyzed P28 striatum following E14.5 induction of Ascl1-CreER;Ai14 mice and detected only M cells (Figures 3Q, 3U and 3V). These results indicate that late-phase Ascl1+ bIPs are fate-restricted to generate M cells.
As Dlx1 may act downstream and/or in parallel to Ascl1 in the LGE (Long et al., 2009), we also performed systematic fate mapping using the Dlx1-CreER;Ai14 mice throughout LGE neurogenesis (Figures 3N–3P and 5). Similar to the early-phase Dlx1+ cells’ positions, late phase (E13.5 to 17.5), pulse-chase of Dlx1+ cells labeled post-mitotic neurons in the MZ, but excluded RGs and aIPs in the VZ (Figures 3N–3P and S3G-S3J). In striking contrast to the early-phase pulse-chase results, however, late-phase pulse-chase did prominently label bIPs with no apical end-feet throughout the SVZ (Figures 3N–3P). Thus, the recruitment of Dlx1 transcription factor distinguished putative bIPs from aIPs. Furthermore, when assaying fate-mapped SPNs in the mature striatum, we found that late-phase Dlx1+ bIPs also specifically produced SPNs of the matrix compartment (Figures 3R–3V). These M cells formed an expansive and dense field of SPNs within which unlabeled striosomes were embedded (Figures 3R–3T). Over 98% of these IP progeny were co-labeled with the matrix marker CDGI, but not with the striosome marker MOR1. These SPNs projected primarily to the non-dopamine-containing SNr, largely sparing the dopaminergic SNc (Figures 4B, 4D and S4C; Supplemental Movie 5). Together with evidence from analysis of Ascl1 and Dlx1 mutant mice (Yun et al., 2002), these results suggest that during the late phase, Dlxl likely acts down-stream and/or in coordination with Ascl1 in bIPs to promote M cell production, in addition to its role in postmitotic SPNs.
The Sequential Production of Striosomal and Matrix SPNs is Punctuated by a Sharp Transition Period
As Ascl1 and Dlx1 are expressed in the LGE throughout embryonic neurogenesis and are thought to act sequentially as well as in parallel in regulating striatal patterning (Long et al., 2009; Wang et al., 2013), we extended our analyses by performing an extensive set of fate-mapping experiments with Ascl1- and Dlx1-driver mice in time-steps of 0.5-1.0 day spanning the entire embryonic neurogenic period from E9.5 to E18.5 (Figures 5 and S4). The results (Figure 5) indicated a surprisingly sharp shift in neurogenic programs for the generation of striosomal and matrix SPNs, likely driven by largely distinct S cell-producing aIPs (aIPS) and M cell-producing bIPs (bIPM) pools.
In summary, during an early phase (E9.5-E12.5), Ascl1 is expressed in aIPs and Dlx1 in postmitotic neurons; TM induction of both drivers predominantly labeled S cells (Figures 5A and 5C). This early period was followed by a transition phase in which there was a rapid decline of S cell production starting around E12.5. From E13.5 on, S cell production became negligible, whereas the labeling of M cells from both drivers was strongly initiated at ~E13.5 and rose sharply within ~24 hours to peak levels. In the subsequent late phase, Ascl1 expression shifted from aIPs in the VZ to bIPs in the SVZ, whereas Dlx1 expression emerged specifically in SVZ bIPs in addition to postmitotic SPNs in the MZ. Between E13.5 and E18.5, the Ascl1 driver initially labeled progenitors producing M cells until ~E16.5, before a sharp decline that was correlated with a switch to the generation of glial cells from the LGE (Figures 5A and 5D). By contrast, the Dlx1 driver persisted in marking progenitors in an active phase of M cell production until the end of neurogenesis at ~E18.5. During both the early and transition phases, the onset and progression of S cell production and the switch from S to M cell production occurred earlier in the Ascl1 driver than in the Dlx1 driver, suggesting that Dlx1 might act downstream of Ascl1 in a transcription cascade that regulates neurogenesis and SPN specification. The MZ expression of Dlx1 in the early and late phase neurogenesis is consistent with this notion (Figures 2E and S3G). The divergence of Ascl1 and Dlx1 cell progeny in the late phase, with Ascl1 driving the production of glia and Dlx1 driving production of M cells, suggests that Ascl1 and Dlx1 probably also act in parallel neurogenetic programs during cell type specification.
Collectively, these results support the existence of two distinct and sequential types of LGE IPs that generate the SPNs of the two major neurochemical compartments of the striatum. An early set of aIPS in the VZ generate striosomes with limited neurogenic capacity, whereas a late-activated set of bIPM generate matrix SPNs with expanded capacity. Quantification of the total number of fate-mapped S and M cells across the embryonic neurogenic period indicated that the proportion of S and M cells produced was, respectively, approximately 20% and 80% (Figure 5F), matching the approximate ratio of total S and M cells estimated for the mature striatum (Johnston et al., 1990; Mikula et al., 2009).
Given that both Ascl1 and Dlx1 are expressed not only in the LGE but also in the MGE, which generates both cortical and striatal interneurons (Marin et al., 2000; Zhao et al., 2003), it was possible that striatal interneurons were also labeled by TM induction in these driver lines, and that these contributed to, or obscured, the striosome and matrix patterning that we observed. To address the degree to which MGE-derived GABAergic interneurons were intermixed with fate-mapped SPNs derived from the LGE, we devised an intersectional strategy specifically to fate map Ascl1+ or Dlx1+ progenitors within the MGE (Figure S5). We combined Ascl1-CreER or Dlx1-CreER drivers with a constitutively active Nkx2.1-Flp driver (which targets all Nkx2.1 positive MGE progenitors; He et al., 2016) and the intersectional Ai65 reporter (Madisen et al., 2015) by crossing the corresponding mouse lines. This approach allowed us to restrict fate mapping to Ascl1+ and Dlx1+ progenitors in the MGE and preoptic area, yielding only GABAergic interneurons following TM induction at any given time point (Figure S5). Consistent with previous reports, the overwhelming majority of MGE-derived GABAergic interneurons were born during late embryonic days (~90% from E16.5-E18.5), with very few originating during the period of striosome neurogenesis (Figures S5B-S5D). Similarly, we found parvalbumin-positive and somatostatin-positive interneurons to be scarce within fate-mapped striosomes, but to be located in the matrix compartment when labeled (Figure S6). Cholinergic interneurons were also not observed to be part of Ascl1 and Dlx1 fate-mapped cohorts regardless of induction time, whether from MGE only (intersection) or MGE/LGE (Ai14) (Figure S6). This result suggests that striatal cholinergic interneurons have a distinct progenitor origin from SPNs in the LGE, or that they are generated at a time-point or location not covered by our analysis.
The Temporal and Spatial Sequence of SPN Settlement in the Striatum Include a Final Cohort That Constitutes Peri-Striosomal Rings
Consistent with previous studies (Krushel et al., 1993; van der Kooy and Fishell, 1987; Newman et al., 2015) Hagimoto et al., 2017; we observed strong spatial gradients in the settling of SPNs along both the lateral-medial and caudal-rostral axes according to the time of Ascl1 and Dlx1expression (Figures 5 and S7; Supplemental Movie 6). More specifically, the SPNs of the matrix compartment not only follow such developmental time-dependent gradient distributions, but are further divided into extended mosaics of cell-clusters (“matrisomes”), which become obvious in tracing experiments demonstrating the input-output organization of the matrix (Flaherty and Graybiel, 1993, 1994; Giménez-Amaya and Graybiel, 1991). Although tract-tracing (Alloway et al., 1999) and 2-deoxy-glucose (Brown et al., 2002) analyses suggest that such patchiness occurs in the striatal matrix in rodents, matrisomes are not visible in conventional histological preparations. Nevertheless, we found a striking non-uniformity in the matrix in our fate-maps: a gap around striosomes with SPNs labeled from E15-E17 cohorts (Figure 3S) that were filled in by a peri-striosomal distribution of the last-born E18 cohort of SPNs (Figure 6). These E18-derived SPNs adhered tightly to the borders striosomes (Figures 6A–6E), and were reminiscent of the peri-striosomal rings demarcated at maturity by immunochemical markers and interneuron distributions observed in primates and humans (Faull et al., 1989; Holt et al., 1997; Graybiel and Ragsdale, 1980). It is thought that these peri-striosomal regions may functionally link striosome and matrix compartments (Banghart et al., 2015; Brimblecombe and Cragg, 2015; Miura et al., 2007). We found that peri-striosomal cells were born almost exclusively during the terminal rounds of neurogenesis at E18.5 (Figure 6). These cells accounted for ~65% of total E18.5 fate-mapped SPNs in whole striatum at P28. Regionally they were more enriched in rostral striatum (~80% of fate-mapped SPNs), reflecting the fact that there were also a significant number of caudal matrix SPNs still being generated at this late time, which mainly populated the medial matrisomes bordering the lateral ventricle (Figures 6C–6F). We also noted similar ring-like structures surrounded MOR1-positive patchy zones in the ventral striatum (Figure 6A). Overall, we conclude that M cells generated from E15.5-E18.5 were deployed in a distal to proximal sequence in relation to the striosomes (Figures 6G–6I). As the RFP-labeled axons of the late-born peri-striosomal M cells innervated both pallidal segments and the SNr, the typical pattern of M cell axons, it is highly likely that at least part of the peri-striosomal M cells correspond to the annular compartment as originally designated by Faull et al. (1989).
Direct and Indirect Pathway SPNs in the Striosomes and Matrix Originate Independently within Both aIPS and bIPM
How the neurogenic programs producing striosome-matrix architecture relate to the generation of SPNs of the direct and indirect pathways has long been a major question about basal ganglia organization. Direct and indirect classes of SPNs are distinguished from one another by multiple molecular markers, including their differential expression of D1 or D2 variants of the dopamine receptor (Gerfen et al., 1990; Harrison et al., 1990; Richfield et al., 1989). These markers occur in both striosomes and matrix, but in sharp contrast to the ~1:5 ratio of spatial compartmentalization of striosomal and matrix SPNs, dSPNs and iSPNs are intermixed in both compartments (Fujiyama et al., 2011; Gerfen and Scott Young, 1988; Flaherty and Graybiel, 1994; Banghart et al., 2015; Brimblecombe and Cragg, 2017; Salinas et al., 2016).
To determine whether S- and M-cell fates relate to the acquisition of direct and indirect pathway fates, we distinguished the dSPNs and iSPNs in our genetic fate mapping of the striosome and matrix compartments by relying on the tight correlation between dSPN with D1 receptor expression and iSPN with D2 receptor expression (Gerfen et al., 1990; Harrison et al., 1990; Richfield et al., 1989). We identified these SPNs by fluorescent mRNA double in situ hybridization with D1 and tdTomato probes or with D2 and tdTomato probes performed on brain sections in Ascl1-CreER;Ai14 and Dlx1-CreER;Ai14 mice induced at specific embryonic times. Thus we simultaneously identified neurons of the striosome and matrix compartments in terms of their D1 and D2 expression levels (Figure 7).
The spatial pattern of tdTomato mRNA-expressing S cell clusters was readily distinguishable from the M cell distributions. By contrast, D1 and D2 mRNA signals were evenly distributed in apparently equal numbers intermixed throughout the striatum (Figures 7C and 7D). Among S cells labeled by TM induction of Ascl1-CreER;Ai14 mice at E10.5 and E12.5, approximately equal ratios of D1 and D2 cells were detected, with a slight enrichment of D1 cells (Figure 7I). Similarly, among M cells labeled by TM induction of Ascl1-CreER;Ai14 mice at E15.5, we found equal numbers of D1 and D2 cells (Figure 7I). Among M cells labeled by TM induction of Dlx1-CreER;Ai14 mice at E17.5, again both D1 and D2 cells were detected, but with a significant enrichment of D2 cells (Figure 7I). Together, these results indicate that, at least at the population level, the generation of dSPNs vs. iSPNs is not specified by a program analogous to one that generates striosome vs. matrix compartments. The remarkable conclusion suggested by these results is that the specification of a direct-indirect pathway fate appears to be mainly independent of striosome-matrix fate and manifests within both the aIPS and bIPM progenitor pools, though with a slightly skewed distribution toward a D1-striosome/D2-matrix bias.
DISCUSSION
Two fundamental categories of vertebrate behaviors that enhance fitness encompass action selection, involving evaluation of multiple options, and action evaluation, based on behavioral outcome. In mammals, action selection and evaluation are crucially regulated by basal ganglia circuitry (Amemori et al., 2011; Graybiel, 2008; Hikosaka et al., 2014), through which the cerebral cortex modulates a wide range of volitional and habitual behaviors. As the recipient of multi-modal information represented in different neocortical areas and thalamic nuclei, the striatum is thought to reconfigure and then to forward the resultant signals to basal ganglia output circuits to initiate and maintain appropriate actions while suppressing others, and to evaluate their outcomes (Mink, 1996; Graybiel and Grafton, 2015). To support such sophisticated operations, the circuit architecture of the striatum is subdivided into striosome and matrix compartments, each further containing direct and indirect pathway subdivisions of SPNs as defined by their selective expression of dopamine receptor variants. Here, we have discovered that the basic plan for assembling the striatal circuit architecture has its root in a developmental program embedded in the LGE RGs and their lineage progression, which generate distinct fate-restricted IPs to specify SPNs into striosome and matrix types, upon which an independent and secondary mechanism divides SPNs into the direct and indirect pathways. This set of findings raises the possibility that the deployment of different progenitor types may reflect an evolutionarily conserved mechanism to assemble striosomal and matrix circuits that mediate distinct behavioral categories involving action selection and evaluation (Grillner and Robertson, 2015; Amemori et al., 2011; Graybiel, 1997; Graybiel and Kimura, 1995; Graybiel, 1997; Houk et al., 1994) while the approximately equal ratio of intermixed direct and indirect pathway SPNs within the two compartments could represent a mechanism for more flexible configuration of functional sub-circuits based on specific outcome evaluation and motor outcome (i.e., experience). This view further raises the intriguing and more general possibility that, as in invertebrates in which cell lineage and developmental genetic program shape circuits and behavior (e.g., Harris et al., 2015), a developmental ground plan in mammalian telencephalic neural progenitors may direct the assembly of certain basic forebrain circuit organization underlying high-order behaviors.
Lineage Dependent and Independent Production of Different Types of SPNs
The actions of specific transcription factors, such as Gsx1/2, Ascl1, and Dlx1/2, as well as Notch signaling, mediate cell autonomous and non-autonomous regulation of neurogenesis in the LGE and control ordered production of striatal neurons (Mason, 2005; Yun et al., 2002). It has been suggested that the early LGE contains Gsx1/2+ neuroepithelial (NE) cells that give rise to multiple progenitor states characterized by Ascl1 and Dlx expression (Martín-Ibáñez et al., 2012; Yun et al., 2002). Ascl1+/Dlx1/2− and Ascl1+/Dlx1/2+ progenitors are inferred to emerge in a sequence (Martín-Ibáñez et al., 2012) and to interact through Notch-mediated lateral inhibition to coordinate proliferation and neurogenesis and to regulate orderly production of early- and late-born SPNs (Mason, 2005). Here, by combining cellular resolution genetic fate mapping and neuronal birth dating, we were able to identify distinct progenitor types, to uncover their lineage progression, and to track their neuronal progenies to SPN types. Based on our findings, we propose a working model of the progenitor and lineage mechanisms underlying the production of SPN types and their influence on striatal circuit architecture (Figure 8).
According to this model, a multi-potent radial glial lineage sequentially allocates distinct intermediate progenitor (IP) types to build striosome-matrix compartmentalization, and the division of direct and indirect pathways occurs within each IP type secondary to the compartmental divide. More specifically, the onset of neurogenesis in the VZ of early LGE (~E10) is marked by Tis21 expression in RGs, which self-renew over multiple rounds to generate several types of IPs and all types of striatal SPNs. In the early phase of this lineage progression, RGs may generate neurons directly and/or through aIPs (including SNPs and SAPs) with dynamic ASCL1 expression patterns. Early aIPs (herein termed aIPScells) with sustained high ASCL1 levels are committed to produce striosomal SPNs with no or likely no more than one round of amplification, thus limiting their neurogenic capacity. These aIPSs produce both striosomal dSPNs and iSPNs in an approximately equal ratio. During a brief transition period starting ~E12.5, RGs undergo a rapid shift from generating aIPs to generating bIPs that are fate-restricted to produce M cells (i.e., bIPM). Dlx1 is activated downstream and/or in parallel to Ascl1 in bIPM. A possible rapid depletion of aIPSs may result in a sharp decline of S cell production, while the rapid rise in the production of bIPMs promotes M cell genesis. In the next phase (E14.5-E17.5), Ascl1+ bIPMs first promote M cell production and then switch to the generation of glial cells. By contrast, Dlx1+ bIPMs persist in driving highly prolific M cell production throughout the late phase of LGE neurogenesis. Many bIPMs may undergo one or more rounds of transit-amplifying proliferations before the final neurogenic division, thereby increasing their neurogenic capacity, in addition to the prolific generation of bIPs by RGs through aIPs, and likely proliferation of the RG pool itself. Such mechanisms likely 15 account for the significant numerical expansion of the matrix relative to the striosome compartment. At the final stage (~E18.5), Dlx1+ IPs generate SPNs destined to form peri-striosomal ring structures, of which some subset likely contribute to the annular compartment. Throughout bIPM-mediated neurogenesis, as during the aIPS-mediated period that precedes it, a similar ratio of matrix dSPNs and iSPNs are generated, with a bias toward iSPNs at the end of bIPM neurogenesis.
Our fate-mapping method cannot distinguish direct vs. indirect neurogenesis from Tis21+ RGs. Additionally, the mechanism underlying the shift of RG state from aIPS to bIPM production is unknown and might involve RG intrinsic (e.g., tracking cell division number) or extrinsic (e.g., signaling from increasing number of S cells) processes; transcriptome analysis of early and late RGs may uncover the molecular basis.
The Relation of Striosome-Matrix to Direct-Indirect Pathway SPN Origins
A remarkable finding from our fate-mapping experiments is that dSPNs and iSPNs appear to derive from a different developmental process than that giving rise to the striosome and matrix compartments: both dSPNs and iSPNs derive from both aIPS and bIPM pools and in approximately equal or biased ratios, depending on the embryonic day, throughout the LGE neurogenic window. How could such a consistent pathway division be set up within the developmental process giving rise to striosomes and matrix? One possibility is that when S or M cells are first specified, they are neutral with respect to the D1-D2 distinction, so that these phenotypes are acquired in response to environmental factors or cell-cell interactions. An alternative possibility is that the D1 and D2 fates emerge together with the fundamental distinction into striosome or matrix identities. Thus, it is formally possible that, within the aIPS and bIPM pools, subsets of IPs are further fate-restricted to generate direct vs. indirect pathway S cells, and direct vs. indirect pathway M cells. We also note that a stochastic mechanism (Boije et al., 2014) might be well suited to generate the characteristic ratio of the two pathway SPNs, as stochastic decisions can produce a mosaic of fates within a population of cells, and they can be integrated into developmental programs and directed to yield robust and reproducible outcomes (Boije et al., 2014; Johnston and Desplan, 2010). Such a stochastic mechanism for D1 vs. D2 fate could manifest itself either at the step of IP generation from RGs or at the final neurogenic division of individual IPs. Nearly equal or biased ratios of highly intermixed dSPNs and iSPNs generated by stochastic fate choice from IPs could lay out a template in both striosomes and matrix upon which learning and experience can shape behaviorally relevant functional sub-circuits.
The Significance of Distinct IPs and Their Differential Amplification for the Production of Striosomal and Matrix SPNs
In addition to differences in input-output connectivity and behavioral function, the striosome and matrix compartments differ substantially in size and maintain an approximately 1:4 ratio that is conserved across mammalian species. The mechanisms underlying this size difference between the two compartments is not well understood. Our finding that distinct IP types with different neurogenic capacities are fate restricted for S vs. M cell production provides plausible mechanisms. First, a larger number of bIPsM are generated from RGs during a longer late phase and possibly through late-phase aIPs that further amplify bIPsM numbers. Second, individual bIPsM undergo more rounds of transit amplification compared to early aIPsS. We suggest that aIPS-mediated generation of SPNs in the early phase forms an evolutionarily conserved basic template for evaluative circuits. On the other hand, bIPM-mediated amplification (starting from ~E13) dramatically increases the production of matrix SPNs and could underlie the development of the much larger matrix compartment with potential further diversification such the peri-striosomal rings. The greatly expanded matrix SPN population may provide the basis underlying flexibility for building a large repertoire of basal ganglia circuits contributing to action selection and motor skill learning in parallel with the expansion of the neocortex.
The recognition of the distinction between the developmental plan for striosome-matrix compartmentalization and its relation to the direct-indirect pathway subdivision of SPNs opens the way to a deeper examination of striatal circuitry linking development and behavior. As these two organizational plans are now known to harbor different disease vulnerabilities, for example, of striosomes in mood disorders in Huntington’s disease patients and of D2-bearing striatal neurons in Parkinson’s disease, these findings may have significance for understanding the developmental and circuitry etiology of these clinical disorders (Crittenden and Graybiel, 2016).
EXPERIMENTAL PROCEDURES
Mice
Mouse lines are described in the Supplemental Experimental Procedures. All experiments and procedures are performed according to NIH guidelines and approved by the IACUC at Cold Spring Harbor Laboratory and at Massachusetts Institute of Technology.
Inducible Genetic Fate Mapping
Inducible genetic fate-mapping methods are described in the Supplemental Experimental Procedures.
Tamoxifen Induction and BrdU Birth Dating
To reduce toxicity to pregnant females and enable sparse cellular labeling, pregnant female mice were submitted to gavage with low to medium dose tamoxifen (dissolved in corn oil by gentle agitation for 24 hours), using an 18-gauge 50 mm gavage tool. Doses were in the range of 0.15 mg/30 g body weight (BW) up to 5 mg/30 g BW (5 mg/Kg BW–165 mg/Kg BW), with most experiments performed using doses below 1.5 mg/30 g BW. For BrdU birth dating, pulses of BrdU were intraperitoneally injected into pregnant mice at a dose of 5 mg/100 g BW, spaced 4 or 24 hours after TM induction on E10.
Immunohistochemistry and Microscopy
Immunohistochemistry and microscopy are described in the Supplemental Experimental Procedures.
Double Fluorescent in situ Hybridization
Two-probe fluorescent in situ hybridization was performed on fresh frozen 12 µm thick cryosections of P28 mouse brains using the QuantiGene ViewRNA ISH assay from Affymetrix. Briefly, tissue was treated with formaldehyde fixation, protease digestion, target and label probe incubation, and buffer washing according to the QuantiGene Affymetrix protocol. Tissue was then co-labeled with fluorescent Nissl and imaged using a DeltaVision epifluorescent microscope (details described in Supplemental Experimental Procedures).
Imaging and Serial Reconstruction of Whole Brains
Imaging and methods for serial reconstruction of whole brains are described in the Supplemental Experimental Procedures.
Quantification and Data Analysis
Quantification and data analysis are described in the Supplemental Experimental Procedures.
AUTHOR CONTRIBUTIONS
Z.J.H and A.M.G. conceived and organized the study. S.M.K. and Z.J.H. designed the experiments. S.M.K., P.W. and M.H. designed and generated Tis21-2A-CreER and Nkx2.1-ires-Flp knock-in mouse lines. S.M.K and R.R. conducted primary genetic fate-mapping experiments and embryonic pulse-chase experiments. J.L., L.G.G. and A.M.G. performed comprehensive tamoxifen inductions and histology for each half-day fate-mapping time point and collected brain samples for in situ hybridization experiments. S.M.K. completed fluorescent double in situ experiments. S.M.K., Y.K. and P.O. performed serial two-photon imaging experiments. J.L. and K.M. contributed to immunostaining and imaging of striatal interneuron markers in fate-mapped cells. Z.J.H, A.M.G. and S.M.K. wrote the manuscript.
ACKNOWLEDGMENTS
We are grateful to Sang Yong Kim for help with generation of knock-in mice, to Nour El-Amine for help with imaging of fluorescent double in situ samples, and to Gord Fishell, Songhai Shi, Marcus Stepheson-Jones, Bo Li, and Yasuo Kubota for discussions and invaluable comments on early versions of this manuscript. This work was supported in part by a grant from the Simons Foundation to Z.J.H. and A.M.G., by CSHL Robertson Neuroscience Fund to Z.J.H., by a grant from the CHDI Foundation (A-5552) to A.M.G., and by support from the Saks Kavanaugh Foundation to A.M.G. S.M.K. was supported by NRSA F30 Medical Scientist Predoctoral Fellowship 5F30MH102002-03, and L.G.G. was supported in part by a Postdoctoral Fellowship from the William N. & Bernice E. Bumpus Foundation. Work is also Supported also by NIMH grant 1S10OD021759-01