A Transcription Factor Code Defines Nine Sensory Interneuron Subtypes in the Mechanosensory Area of the Spinal Cord

Interneurons in the dorsal spinal cord process and relay innocuous and nociceptive somatosensory information from cutaneous receptors that sense touch, temperature and pain. These neurons display a well-defined organization with respect to their afferent innervation. Nociceptive afferents innervate lamina I and II, while cutaneous mechanosensory afferents primarily innervate sensory interneurons that are located in lamina III–IV. In this study, we outline a combinatorial transcription factor code that defines nine different inhibitory and excitatory interneuron populations in laminae III–IV of the postnatal cord. This transcription factor code reveals a high degree of molecular diversity in the neurons that make up laminae III–IV, and it lays the foundation for systematically analyzing and manipulating these different neuronal populations to assess their function. In addition, we find that many of the transcription factors that are expressed in the dorsal spinal cord at early postnatal times continue to be expressed in the adult, raising questions about their function in mature neurons and opening the door to their genetic manipulation in adult animals.


Introduction
Interneurons in the dorsal spinal cord receive and process multiple types of cutaneous sensory information, including pain, temperature, pressure and vibration [1][2][3][4][5][6][7]. In addition to relaying cutaneous stimuli, interneurons in the dorsal horn transmit propioceptive information from Group II and III muscle afferents [8]. These cutaneous sensory afferents terminate in the dorsal horn in a modality-specific manner [1,2,7]. Nociceptive information is received primarily in lamina I-II from two different classes of sensory afferent neurons that are distinguished molecularly as peptidergic (CGRP + /TrkA + ) C/Ad fiber afferents and nonpeptidergic (Mrgprd + /IB4 + /Ret + ) C fiber afferents [7]. There are also nociceptive Ad fibers that terminate in lamina I [7]. Low threshold mechanoreceptors (LTMRs) that transduce innocuous cutaneous mechanosensory information innervate first order sensory interneurons that are located between inner lamina II (IIi) and lamina IV [8]. As a general rule, C-fiber LTMRs primarily project to lamina II, Ad-fiber LTMRs project to laminae IIi and III, while Ab-fiber RA-LTMRs project mainly to laminae III-IV [7,8]. Proprioceptive information in the dorsal spinal cord is mainly processed by neurons in laminae IV-VI, although many proprioceptors project to more ventral regions of the spinal cord where they innervate premotor interneurons and motor neurons [1,[9][10][11].
Despite the importance of the dorsal spinal cord for the reception and transduction of cutaneous mechanosensory stimuli, we know very little about the neuronal composition of the central circuits that gate and transmit this information. Efforts to probe the organization of these circuits have been hampered by their complexity, and by an inability to molecularly define discrete populations of sensory neurons and ascribe functions to them. Recently, a number of developmentally-regulated transcription factors that are expressed in the developing dorsal horn have been identified [11][12][13][14][15][16] that provide an entry point for identifying the sensory interneuron cell types that play essential roles in processing and transducing cutaneous somatosensory information. Using a battery of transcription factors that are expressed at late embryological and early postnatal stages, we have begun to probe the molecular diversity of interneurons in laminae III-IV, which is primarily innervated by cutaneous mechanoreceptors. The resultant analysis of multiple transcription factors in combination with Pax2, Gbx1 and Lmx1b, which are more broadly expressed in the dorsal spinal cord, has allowed us to identify nine molecularly-distinct interneuron populations in lamina III-IV at postnatal and adult stages. More importantly, the systematic identification of nine molecularly-defined sensory interneuron cell types in lamina III-IV has set the stage for functionally dissecting mechanosensory circuits in lamina III-IV using genetic and molecular approaches similar to those employed for studying central pattern generator (CPG) networks (for recent reviews see [17][18][19]). Consequently, we can now: 1) examine the role that specific neural populations play in transducing the sensation of touch, 2) determine the contribution that cutaneous stimuli make to the dynamic control of movement, and 3) further our understanding of how somatosensory information is coded by spinal cord interneurons.

Animals
All protocols for animal experiments were approved by the IACUC of the Salk Institute for Biological Studies and follow the NIH guidelines for animal use. The mouse lines used in this study have been described previously: Pax2-Cre [20]; R26 floxstop-Tomato (Ai14) [21]; MafB-GFP [22]; GAD67-GFP [23]; Lmx1b knockout [24]; RORa-IRES-Cre [25]. All mice were genotyped by PCR using allele-specific primers for each strain. For timed pregnancies, midday on the day of the vaginal plug was designated as embryonic day (E) 0.5. The day of birth was designated as P0. Tissue Preparation and Immunohistochemistry P0-adult mice were euthanized and perfused with 4% paraformaldehyde in PBS (PF). Their spinal cords were then post-fixed for 30-60 mins in 4% PF at 4uC (P0) or at room temperature (adult). Spinal cords were rinsed and cryoprotected in 20% sucrose in PBS (4uC) prior to embedding in OCT (Tissue-Tek). Immunostaining of frozen spinal sections was performed by incubating 20 mm thick sections with primary antibodies, which were then detected using species-specific secondary antibodies conjugated with Cy2, Cy3 and Cy5 (Jackson Laboratories) or FITC (Invitrogen). Three-color images were captured using either an Axioskop 2 Mot Plus microscope or a Zeiss LSM510 Laser Scanning Confocal Microscope. AxioVision and Adobe Photoshop software was used for image analysis, data processing and presentation.

Identification of Transcription Factors Expressed in the Mechanosensory Area of the Dorsal Horn at Postnatal and Adult Stages
We used a battery of antibodies, in combination with previously characterized mouse reporter lines, to map the expression of multiple transcription factors in the postnatal dorsal horn and test whether they are expressed in discrete populations of laminae III-IV sensory interneurons. This initial survey identified a number of transcription factors with known roles in neuronal specification and differentiation, all of which are expressed in the dorsal spinal cord at late embryonic and postnatal stages ( Fig. 1; data not shown). We focused our analysis on thoracic and lumbar spinal cord levels, as there was no discernable difference in the expression patterns of these transcription factors along the anterior-posterior (A-P) axis at lower spinal cord levels. Transcription factor expression was analyzed at the following ages: P0, P3, P4, P7, P10, P27, and in the adult.
Two broad sets of transcription factors were identified. The first, which includes Lbx1, Lmx1b, Tlx3, Pax2 and Gbx1, is comprised of transcription factors that display relatively broad patterns of expression in the dorsal horn ( Fig. 1A-E). During embryogenesis, these transcription factors are expressed broadly in the developing spinal cord [11,12,14,15,27,28], whereas at postnatal times, we find that they are enriched in the dorsal horn, including laminae III-IV ( Fig. 1K-N). The second set of transcription factors that we identified displays a more restricted pattern of expression in dorsal sensory interneurons. This group includes the nuclear orphan receptors RORa and RORb, and the large Maf proteins, MafA, MafB and c-Maf. All five transcription factors displayed cell typespecific expression in the dorsal horn at postnatal stages ( Fig RORb, on the other hand, displays a more restricted pattern of expression throughout development (data not shown). In the postnatal and adult cord, RORb is restricted to laminae III-IV, with the exception of a few cells that are located in lamina I ( Fig. 1G; Fig. S2).
Lmx1b is expressed at high levels in laminae I-III, with lower levels of expression in lamina IV (Fig. 1L). Tlx3 is principally expressed in laminae I-II, although we did detect Tlx3 + cells in laminae III (Fig. 1J, M). Pax2-expressing cells were more broadly distributed within the dorsal horn (Fig. 1D), but were present in lower numbers in the ventral horn. The ventral Pax2 + cells, together with the dorsal Pax2 + cells, are likely to be inhibitory interneurons, due to their expression of multiple inhibitory neuron markers at early embryonic times [17,[27][28][29].
In summary, we have identified a cohort of transcription factors that are expressed in sensory interneurons within lamina III-IV, which is the primary recipient region for innocuous mechanosensory afferents. In view of the demonstrated roles that these transcription factors play in regulating neuronal cell specification, differentiation and cell physiology [11,12,14,15,[36][37][38][39][40][41][42][43], it is highly likely that they have important roles in controlling the physiology of dorsal sensory interneurons gating cutaneous mechanosensory stimuli.   Table 2 for further details. doi:10.1371/journal.pone.0077928.g004 [11,12,28,29]. To address this question, we used glutamic acid decarboxylase 67-green fluorescence protein (GAD67-GFP) knockin mice [23] to mark and trace inhibitory interneurons. Lmx1b was not detected in neurons that express GFP at any postnatal time analyzed, indicating the Lmx1b cells are not GABAergic inhibitory interneurons (Table 1, data not shown). When we analyzed Pax2-Cre; R26 floxstop-lacZ ; Gad67-GFP mice, most, if not all, Pax2-derived b-galactosidase + (b-gal + ) cells in laminae III-IV expressed GFP, demonstrating that they are indeed inhibitory neurons ( Fig. 2A,A9). Gbx1 also showed strong co-localization with the GAD67-GFP reporter (Fig. 2B,B9), and a large number of these Gbx1 + cells expressed b-gal (Fig. 3D-E9). We also detected Pax2 + and Gbx1 + cells that do not express GFP in the postnatal cord ( Fig. 2A-B9). The presence of these GAD67-GFP-negative cells in the postnatal cord most likely reflects the down-regulation of GAD67 at postnatal times, which has been noted in other studies [23,28]. Conversely, there are GAD67-GFP + neurons at in the postnatal cord that do not express b-gal or Gbx1 ( Fig. 2A,B). In Pax2-Cre; R26 floxstop-lacZ mice, the neurons that continue to express Pax2 were found to represent only subset of the Pax2-Cre + (b-gal + ) cells in dorsal horn (Fig. 2C,C9). This is again due to the down-regulation of Pax2 (and Gbx1) in the postnatal dorsal horn [28,30], since at earlier times Pax2 is expressed in all Pax2-Cre marked neurons (Fig. S3).
The Pax2-Cre; R26 floxstop-lacZ reporter mouse was then used to assess whether Lmx1b co-localizes with GFP in Pax2-derived ''inhibitory'' neurons. No overlap between Lmx1b and b-gal expression was noted at P1 (Fig. 2G), which is consistent with our observation that Lmx1b is excluded from cells that belong to the Pax2 (dIL A ) lineage (Fig. S3). This finding demonstrates that Pax2-Cre-derived neurons do not express the excitatory marker Lmx1b, and are thus unlikely to be glutamatergic interneurons. We also confirmed that Lmx1b does not co-localize with Gbx1 (Fig. 2F), indicating that Lmx1b and Gbx1 mark two separate cell populations in the dorsal horn. The majority of laminae III-IV cells that express either Pax2 or Gbx1, express both factors, although there are a small number of cells that express Pax2 and Gbx1 alone (Fig. 2E). Taken together, our data demonstrate that Lmx1b and Pax2/Gbx1 are specific postnatal markers of excitatory and inhibitory neurons, respectively.
We then analyzed GAD67-GFP mice at P3, P7 and P10 and  Table 1). These analyses revealed that Lbx1, RORa, RORb, MafB and c-Maf are present in mixed populations of inhibitory and excitatory neurons. At postnatal times, Lbx1 and RORa are predominantly expressed in Lmx1b + excitatory neurons (Fig. 3A-D), although a small, but significant fraction of the Lbx1 + interneurons in GAD67-GFP mice are GFP + (Fig. 4A-B; data not shown). This expression of Lbx1 in inhibitory neurons was confirmed by double immunostaining experiments with antibodies to Gbx1 and Lbx1 (Fig. 4D, asterisk).
Our results suggest that most, if not, all of the MafA + neurons in lamina III and lamina IV are excitatory glutamatergic neurons, as more than 90% of these MafA + cells co-express Lmx1b (Fig. 3K-L, Table 1). Furthermore, MafA does not co-localize with GFP in GAD67-GFP mice at P3, P7 and P10 (Fig. 4Q-R), nor is it expressed together with Gbx1 at these times ( Fig. 4S-T     Among the five inhibitory neuron populations, there is one that does not express Gbx1. These cells can be seen in the GAD67-GFP spinal cord, where there are a number of GFP + cells in laminae III-IV that do not express Gbx1 (Fig. 2B,B9: Table 2, column 5). The neurons that express Gbx1 can be further subdivided into four populations: one that expresses RORb (  (Fig. 4G-H) do not express MafB (Fig. 5A), in so far as we could not find RORb + /Gbx1 + cells that express GFP in MafB-GFP mice (Fig. 5A-A9; Table 2, column 1). It should be noted that the GFP reporter only labels inhibitory MafB + neurons in these mice ( Fig. 6; data not shown). This is probably due to the loss of an enhancer element in the MafB-GFP knock-in allele that directs GFP expression in excitatory neurons [22,32]. A number of the MafB + (GFP + ) neurons in the MafB-GFP spinal cord were found to express Gbx1 alone (Fig. 5A-A9: Table 2, column 2-3). Some of these cells also express RORa, as there is small population of RORa + (Tomato + )/Gbx1 + cells that express MafB in the RORa-Cre; R26 floxstop-Tomato mice ( Fig. 5B; Table 2, column 2). There is also a population of Gbx1 + /MafB + neurons in the RORa-Cre; R26 floxstop-Tomato mice that do not express RORa (Tomato), which represents the third population of Gbx1-expressing neurons ( Fig. 5B; Table 2, column 3). Finally, in MafB-GFP reporter mice, there are Gbx1-expressing neurons that do not express MafB (GFP) or RORb. These Gbx1 + /MafB 2/ RORb 2 cells constitute the fourth population of Gbx1 + neurons (blue cells in Fig. 5A; Table 2, column 4). We have also found a significant fraction of Gbx1 + cells that express c-Maf (Fig. 5E). These cells are not part of the Gbx1 + /MafB + or Gbx1 + /RORb + populations, as GFP + / c-Maf + cells are rarely, if ever, detected in the MafB-GFP spinal cord ( Fig. 5C and data not shown), and Lmx1b 2/ RORb + inhibitory neurons do not express c-Maf (Fig. 5D-D0; Table 2, columns 1 and 4).

GAD67-GFP
Four different populations of excitatory neurons were identified in lamina III-IV on the basis of Lmx1b and MafA expression, three that express MafA (Table 2, column 6-8), and one that expresses Lmx1b, but not MafA (Table 2, column 9; Figs. 3K-L and 5K9). Interestingly, all of MafA + neurons express Lbx1 (Fig. 5J), although in some instances only very weakly (data not shown). Within the MafA population, one subpopulation expresses RORb. The MafA + /RORb + cells make up approximately half of all RORb + neurons in lamina III-IV ( Fig. 5F; Table 2, column 6). Since RORb is also co-expressed with Tomato in RORa-Cre; Rosa26 floxstop-Tomato mice (KG and SB, unpublished observations), we conclude that this subset of MafA + excitatory neurons most likely expresses a combination of both RORa and RORb ( Table 2, Table 2, column 6). c-Maf rarely, if ever, colocalizes with GFP in MafB-GFP mice where the GFP reporter selectively labels inhibitory neurons (Fig. 5C), leading us to conclude that many of the MafB neurons in lamina III-IV are excitatory c-Maf + interneurons (Table 2, column 7). Our results also demonstrate that there are twice as many MafA + /c-Maf + neurons in lamina III-IV as compared to MafA + /RORb + cells. This means that approximately 50% of the MafA + /c-Maf + neurons in lamina III-IV are RORb-negative ( Table 2, column  7). Finally, we have found a small population of MafA + neurons in lamina III-IV that do not express c-Maf ( Fig. 5I; Table 2, column 8).

Discussion
In this study, we describe the identification of nine different populations of postnatal neurons that are principally located in laminae III-IV, the main area for processing cutaneous mechanical stimuli in the spinal cord. The classification of sensory interneurons in lamina III and IV was based on a combinatorial transcription factor code comprising of developmental factors known to regulate cell fate specification in the nervous system. Our findings demonstrate an unanticipated level of diversity in the interneuron populations that are located in regions of the spinal cord receiving low-threshold cutaneous mechanosensory inputs. A more limited diversity was suggested by previous electrophysiological studies in vitro [44][45][46][47][48]. In characterizing neurons in laminae III-IV by their mechanoreceptive afferent fiber input and their intrinsic discharge properties, Schneider [47,48] identified four groups of cells: phasic, delayed-firing and tonic, where the tonic population is comprised of two different groups. In a similar manner, Hochman et al. [44] divided laminae III-V neurons into four categories based on their firing properties following intracellular current injection: single spike, phasic firing, repetitive firing, and delayed firing. Morphological studies have also identified differences in terms of cell/dendrite shape and axon morphology (see [49,50], and references there in). Cajal described several types of neurons in laminae II-III [51]. In general, lamina III-IV contains neurons of varying sizes and shapes: rounded, slightly elongated, or spindle shaped cells. There is a lower density of these cells in lamina IV, which also contains large cells. The principal neurons of the susbstantia gelatinosa are small neurons with short axons, which can be classified as central cells or Golgi type II neurons. In addition, there are two larger cell types, stalked cells with rounded soma, and islet cells. However, many neurons do not fit into these neat morphological categories, with lamina IV containing medium sized neurons and larger pyramidal type neurons [1,2,51,52]. The dorsal horn also contains a number of projection neurons, the most prominent of these being spinothalamic/spinoparabrachial neurons in lamina I, III and IV, and the dorsal spinocerebellar neurons that are localized ventral-medially in Clarke's column [1,2,7,53]. The Gbx1 + and Pax2 + inhibitory interneuron cell types that we have identified are unlikely to be projection neurons, as spinal projection neurons are primarily glutamatergic. Moreover, the majority of Lmx1b + and Tlx3 + cells in the dorsal horn are likely to be local circuit interneurons [11,12,14].
More recently, multiple subpopulations of dorsal horn glutamatergic and GABAergic neurons have been identified that express various neuropeptides and calcium binding proteins [29,[54][55][56][57]. Subsets of GABAergic neurons express the Ca2 + binding protein parvalbumin, as well as the neuropeptide transmitters neuropeptide Y (NPY), enkephalin, galanin, glycine and thyrotropin-releasing hormone. There are also small populations of GABAergic neurons that express choline acetyltransferase (ChAT) or nitric oxide synthase (NOS). Glutamatergic interneurons express cholescystokinin (CCK), somatostatin and neurotensin [29,[54][55][56][57]. There is also a subset of enkephalin-positive neurons that are also glutamatergic. The correlation between classifying neurons according to their specific transcription factor profiles (this study) and cell types that have been subdivided according to their morphology or neurotransmitter/electrophysiological properties remains to be determined. Defining these relationships would go a long way toward identifying the functional elements of mechanosensory circuitry in the dorsal horn.
Interestingly, most of the transcription factors that are expressed in laminae III-IV do not label a single or homogeneous population of neurons. Instead, Lbx1, RORb, MafB and c-Maf are expressed in both excitatory and inhibitory neurons. To date, MafA is the only marker that is restricted to excitatory neurons, and even then, it is expressed in three molecularly-distinct populations of excitatory neurons. Somewhat surprisingly, Hu et al. [58] have reported that MafA largely co-localizes with Pax2, whereas our data show that MafA is a specific marker of postnatal excitatory neurons. MafA does not co-localize with GFP in GAD67-GFP mice or with Gbx1 at any of the postnatal stages we investigated. Furthermore, MafA is completely lost in Lmx1b mutant mice that express a normal complement of inhibitory gene markers, including Pax2 (Fig. S4). This finding coupled with the observation that MafA is not reduced in Ptf1a mutant mice [58], when Ptf1a is known to be required for GABAergic neuron differentiation [29,59], strongly argues against the expression of MafA in laminae III-IV inhibitory neurons.
While many of the transcription factors analyzed in this study are restricted to the dorsal horn at postnatal times, their expression patterns in the embryonic cord are often broader and encompass neurons that settle in the intermediate and ventral regions of the spinal cord [60][61][62][63][64]. The one exception is RORb. RORb is expressed in the dorsal horn throughout embryogenesis ( Fig. 1; MDB and MG, unpublished data). MafB, for example, is expressed in differentiating Renshaw cells that are derived from ventral p1 progenitors [32]. MafB is also expressed in motor neurons [32]. c-Maf is expressed at E12.5 in dI1 and dI3 neurons in the dorsal horn [31], which are glutamatergic projection neurons that migrate and settle at more ventral locales in the spinal cord. Taken together, these data make it highly unlikely that any single transcription factor specifies cell type in the dorsal horn. They instead point to neuronal cell identity in the dorsal spinal cord being determined by the combinatorial activities of multiple transcription factors.
Although Lbx1, Lmx1b and Tlx3 transcription factors all have essential roles in neuronal specification and differentiation during the two waves of neurogenesis that give rise to dorsal horn interneurons [11,12,14,15], they continue to be expressed in subsets of lamina III-IV neurons in the adult when neural differentiation has ceased. The functional importance of this persistent expression is not known. One possibility is that these transcription factors are important for maintaining the identity and mature phenotype of sensory interneurons. For example, Lhx1 and Lhx5 are required to maintain Pax2 expression in mature GABAergic neurons [28]. The maintenance of these factors along with Gbx1, MafB, MafA and c-Maf may also be important for the reorganization of cutaneous sensory afferent inputs to the dorsal horn that occurs during the early postnatal period [3,65]. For example, Lmx1b is known to play a role in motor neuron axon guidance [66,67], and it might similarly control axon guidance and remodeling in the dorsal horn. In summary, this study defines a novel transcription factor code for sensory interneurons in lamina III-IV. These first order sensory neurons are the targets of low-threshold cutaneous mechanoreceptors, and their characterization provides a foundation for future experiments to determine how sensory neurons in the dorsal horn encode cutaneous tactile information. and Lmx1b following Pax2-Cre-mediated recombination. Note the near complete overlap in Pax2 and nuclear GFP expression at E16.5 (A-B), whereas nuclear GFP expression is completely excluded from Lmx1b + excitatory neurons (C-D). (TIF) Figure S4 MafA expression in the dorsal horn of Lmx1b mutant mice. Control (C1 and C2) and Lmx1b mutant animals (M1 and M2) were analyzed at P0. Lmx1b is expressed in control mice (A and B), but not in Lmx1b mutant mice (C and D). Expression of MafA in the dorsal spinal cord (E and F) is is also lost in the Lmx1b mutant mice (G and H). Pax2 expression in inhibitory neurons is maintained in Lmx1b mutant mice (K and L) in a pattern that is comparable to control mice (I and J). (TIF)