RESEARCH ARTICLE

Muscle-selective RUNX3 dependence of sensorimotor circuitdevelopmentYiqiao Wang1, Haohao Wu1, Pavel Zelenin1, Paula Fontanet1, Simone Wanderoy1, Charles Petitpre1,Glenda Comai2, Carmelo Bellardita3, Yongtao Xue-Franzen1, Rosa-Eva Huettl4,*, Andrea B. Huber4,‡,Shahragim Tajbakhsh2, Ole Kiehn1,3, Patrik Ernfors5, Tatiana G. Deliagina1, François Lallemend1,6,§,¶ andSaida Hadjab1,§,¶

ABSTRACTThe control of all our motor outputs requires constant monitoring byproprioceptive sensory neurons (PSNs) that convey continuousmuscle sensory inputs to the spinal motor network. Yet themolecular programs that control the establishment of thissensorimotor circuit remain largely unknown. The transcriptionfactor RUNX3 is essential for the early steps of PSNsdifferentiation, making it difficult to study its role during later aspectsof PSNs specification. Here, we conditionally inactivate Runx3 inPSNs after peripheral innervation and identify that RUNX3 isnecessary for maintenance of cell identity of only a subgroup ofPSNs, without discernable cell death. RUNX3 also controls thesensorimotor connection between PSNs and motor neurons at limblevel, with muscle-by-muscle variable sensitivities to the loss ofRunx3 that correlate with levels of RUNX3 in PSNs. Finally, we findthat muscles and neurotrophin 3 signaling are necessary formaintenance of RUNX3 expression in PSNs. Hence, atranscriptional regulator that is crucial for specifying a generic PSNtype identity after neurogenesis is later regulated by target muscle-derived signals to contribute to the specialized aspects of thesensorimotor connection selectivity.

KEY WORDS: Sensory system, Sensorimotor circuit, Dorsal rootganglia, Neuronal specification, Neurotrophins

INTRODUCTIONThe neuromuscular circuitry that controls all body movements relieson constant sensory feedback from the periphery to coordinate its

commands to hundreds of muscles. The sensory components of thisfeedback are the proprioceptive sensory neurons (PSNs) of thedorsal root ganglia (DRG), which convey information fromindividual muscles to specific neuron groups in the spinal cord.Previous studies have indicated that the basic design of thissensorimotor circuit is already established at birth (Mears and Frank,1997), and that its construction is largely independent of patternedneuronal activity (Frank, 1990; Mendelson and Frank, 1991;Mendelsohn et al., 2015), implying differential recruitment ofspecific molecular pathways during the establishment of thesensorimotor connections.

PSNs are identified by their specific co-expression of the runtrelated transcription factor RUNX3, tropomyosin receptor kinase C(TRKC, receptor for neurotrophin 3, NT3; also known as Ntrk3),the ETS transcription factor ER81, parvalbumin (PV) and thevesicular glutamate transporter 1 (VGLUT1) (Oliveira et al., 2003;Lallemend and Ernfors, 2012). Peripherally, they terminate inmuscles and innervate the Golgi tendon organ (GTO) (Ib PSNs) andmuscle spindles (MSs) (Ia and II PSNs). Centrally, their axonsconnect with distinct classes of interneurons and α-motor neurons(MNs) in the deep dorsal and the ventral horn of the spinal cord(Lallemend and Ernfors, 2012). Only Ia proprioceptive afferentsmake monosynaptic connections with MNs, establishing thesensorimotor reflex arc. This circuit is very selective as Iaafferents connect to MNs supplying the same muscles and avoidmaking connections with MNs commanding antagonistic muscles(Eccles et al., 1957; Frank and Mendelson, 1990). Most studiesaimed at understanding the molecular programs that control thedevelopment of the motor circuits in relation to the anatomicalidentity of the peripheral connections have focused on the spinalMN specification (Dasen, 2009; Dasen et al., 2005; Price et al.,2002). Spinal MNs innervating the limbs are organized into distinctanatomical columns, the identity of which is tightly controlled byintrinsic genetic programs during early stages of MNs development,before they project peripherally into the limb, and which areessential for the construction of the stereotypic connections withindividual muscles (Dasen, 2009; Dasen et al., 2005; Price et al.,2002; De Marco Garcia and Jessell, 2008). Unlike MNs, PSNsinnervating a specific muscle are scattered in a mosaic fashionthroughout multiple DRGs (Honig, 1982; Honig et al., 1998), anddo not seem to possess any specific identity other than a genericexpression of genes common to all PSNs before innervating theirmuscle target (Wu et al., 2019). PSNs instead would acquiresubclass identities through extrinsic, presumably target-derived,signals days after their neurogenesis (Wu et al., 2019; Poliak et al.,2016). Hence, genetic manipulation of signaling pathways ortranscription factors affecting the afferent outgrowth and muscletargeting of PSNs often results in a lack of sensorimotor connectionsReceived 20 June 2019; Accepted 17 September 2019

1Department of Neuroscience, Karolinska Institutet, Stockholm 17177, Sweden.2Department of Developmental and Stem Cell Biology, Institut Pasteur, CNRSUMR3738, Paris 75015, France. 3Department of Neuroscience, University ofCopenhagen, Copenhagen 2200, Denmark. 4Helmholtz Zentrum Munchen,German Research Center for Environmental Health, Institute of DevelopmentalGenetics, Neuherberg 85764, Germany. 5Unit of Molecular Neurobiology,Department of Medical Biochemistry and Biophysics, Karolinska Institutet,Stockholm 17177, Sweden. 6Ming Wai Lau Centre for Reparative Medicine,Stockholm node, Karolinska Institutet, Stockholm 17177, Sweden.*Present address: Department of Stress Neurobiology and Neurogenetics, Max-Planck-Institute of Psychiatry, Munich 80804, Germany. ‡Present address: ETHZurich, Department of Biosystems Science and Engineering, Basel 4058,Switzerland.§Co-senior authors

¶Authors for correspondence (francois.lallemend@ki.se, saida.hadjab@ki.se)

F.L., 0000-0001-5484-0011; S.H., 0000-0001-7953-8396

This is an Open Access article distributed under the terms of the Creative Commons AttributionLicense (https://creativecommons.org/licenses/by/4.0), which permits unrestricted use,distribution and reproduction in any medium provided that the original work is properly attributed.

1

© 2019. Published by The Company of Biologists Ltd | Development (2019) 146, dev181750. doi:10.1242/dev.181750

DEVELO

PM

ENT

(Lallemend et al., 2012; Levanon et al., 2002; De Nooij et al., 2013;Patel et al., 2003). For example, deletion of RUNX3, which drivesthe specification of PSNs and is associated with a complete absenceof muscle proprioceptive innervation (Lallemend et al., 2012),results in a large deficit of central innervation (Nakamura et al.,2008). Similarly, without NT3-TRKC signaling, PSNs fail toinnervate their peripheral muscle targets and their centralprojections do not extend further into the ventral horn of thespinal cord (Patel et al., 2003). Earlier studies in chick and a morerecent study in mice provide strong support for a role of the target inassigning muscle-specific identities of PSNs that are likely to play arole in the establishment of specific patterns of central sensorimotorconnections (Poliak et al., 2016; Wenner and Frank, 1995).However, with the exception of a role for Sema3E-PlexinD1signaling in gating one specialized aspect of sensorimotorconnectivity (Fukuhara et al., 2013; Pecho-Vrieseling et al.,2009), little is known about the molecular mechanisms requiredfor the establishment of sensorimotor connections between selectivePSNs and the central motor network. Here, using an experimentalstrategy that depletes RUNX3 expression after peripheralinnervation, we uncovered a transcriptional link betweenperipheral muscle target, likely involving NT3 levels, andselective sensorimotor connectivity during late fetal stages.

RESULTSRUNX3 is essential for the maintenance of cell identity of asubgroup of PSNsTo investigate the role of RUNX3 on the development of PSNs andof the sensorimotor circuits independently of its function onneuronal survival and early aspects of peripheral innervation –which end at E12.5 and E14.0, respectively (Fig. 1A) (Fariñas et al.,1998; Tourtellotte et al., 2001) – we first generated PVCre;Runx3fl/fl

mice, with loxP-flanked Runx3 alleles and Cre expressed inPV-expressing neurons. The PVCre mouse line inducedrecombination in a large majority of PV+ and RUNX3+ PSNs atE16.5 when analyzing brachial DRG from PVCre;R26tdTOM mice(Fig. S1A,B). However, in PVCre;Runx3fl/fl mice, RUNX3 deletionwas observed only from birth, with a modest 20% reduction inRUNX3 expression at P0 (Fig. S1C,D), when the sensorimotorcircuit is already established (Mears and Frank, 1997). We thusdecided to generate AdvCre;Runx3fl/fl mice, with Cre expressedunder the control of the advillin gene (Zhou et al., 2010). Comparedwith the PVCre driver line, the AdvCre line induced recombination inall PSNs between E13.5 and E15.5, and a cross with Runx3fl/fl miceresulted in a near-complete absence of RUNX3 expression in DRGat E15.5 (Fig. 1B,C; Fig. S1E,F). Together, this confirms the use ofAdvCre to delete Runx3 in PSNs just after their peripheralinnervation, and at the time they are growing their axonscentrally within the spinal cord to reach their target. In P0AdvCre;Runx3fl/fl mice, focusing our analysis on the brachialsegments, the number of NF200+ DRG neurons (NF200 labels allmyelinated DRG neurons, including PSNs) (Usoskin et al., 2015),and that of RUNX1+ DRG neurons, which represent a largeproportion of the unmyelinated DRG neurons at birth (Lallemendand Ernfors, 2012; Gascon et al., 2010), were similar betweenmutants and control animals (Fig. 1D,E). Hence, the absence ofRUNX3 after peripheral innervation does not affect neuronalsurvival at this stage. However, the expression of factorsnecessary for the proper development and function of PSNs,such as ER81 and TRKC, and of the marker PV were alldownregulated in P0 AdvCre;Runx3fl/fl DRG compared withRunx3fl/fl control mice (Fig. 1F-K). These results show that

expression of RUNX3 in late embryonic DRG neurons isnecessary for maintaining cell identity of a subgroup of PSNs.

RUNX3 regulates development of central projection ofsubgroups of PSNsFrom E13.5 to E15.5, the central afferents of PSNs enter the dorsalpart of the spinal cord and send axons toward the MNs, which theymake contact with at ∼E17.5 (Mears and Frank, 1997; Ozaki andSnider, 1997). The development of the proprioceptive axonal inputsin the spinal cord has been suggested to involve NT3 and thetranscription factors ER81 and RUNX3, with the deletion of any ofthem leading to severe deficits in the central projection pattern ofPSNs (Patel et al., 2003; Nakamura et al., 2008; Arber et al., 2000).Yet those three factors are necessary for the peripheral outgrowth ofPSNs and for their survival (De Nooij et al., 2013; Lallemend et al.,2012; Patel et al., 2003), preventing any study of their directfunction on sensorimotor connectivity using null mutant mice.Here, we have explored whether the loss of RUNX3 in PSNs fromE15 could perturb the formation of the sensorimotor connectionsbetween PSNs and MNs of the ventral spinal cord (phase iii,Fig. 1A). To this end, we analyzed the distribution of VGLUT1+

sensory bouton contacts with CHAT+ (choline acetyl transferase)MNs at P0, which reflects direct excitatory inputs from Ia PSNs onMNs (Oliveira et al., 2003; Pecho-Vrieseling et al., 2009). Theinnervation territory of PSNs in the spinal cord was divided intothree distinct compartments (Fig. 2A): a ventro-medial (M) and aventro-lateral (L) region corresponding presumably to axial andhypaxial muscle-derived Ia projections, respectively; and anintermediate zone (IZ), where most GTOs and type II MSafferents project (De Nooij et al., 2013; Lallemend andErnfors, 2012). In AdvCre;Runx3fl/fl and Runx3fl/fl mutant pups,VGLUT1 expression levels appeared unchanged in PSNs cellbodies (Fig. 2B,C). We observed a general decrease in the density ofVGLUT1 labeling in these three regions at all brachial levels in themutant (here shown for C5 and C8, Fig. 2D-F), with the largestdefect found in the lateral and medial part of the ventral horn, whichcorresponds to the innervation territory of MNs (scheme in Fig. 1A;Fig. 2F). However, no defect was visible at the thoracic level (Fig.S2A). These results contrasted with the phenotype observed in theRunx3−/−;Bax−/− embryos in which, in spite of the absence of sensoryneuron cell death due to null mutation of the pro-apototic gene Bax(Deckwerth et al., 1996), their central afferents did not grow beyondthe dorsal aspect of the spinal cord, as revealed by immunostainingfor peripherin at E15.5 and VGLUT1 at E18.5 (Fig. 2G,H). Takentogether, our results indicate that PSNs at limb levels requireRUNX3 expression to project correctly into the ventral spinal cord,independently of its role on the outgrowth dynamic of theirperipheral projections. Nevertheless, some proprioceptive afferentsstill terminate both at the intermediate spinal cord and in the MNarea in the absence of RUNX3. We thus hypothesized that RUNX3might regulate the central connectivity of selective PSNs.

Development ofmuscle-selective sensorimotor connectionsthrough RUNX3 activitiesPSNs innervating specific skeletal muscles project centrally to andmake contact with defined pools of MNs innervating the samemuscles (Mears and Frank, 1997; Eccles et al., 1957; Dasen, 2009;Surmeli et al., 2011). This connection rule forms the structural basisof the muscle stretch reflex circuit. To explore whether the absenceof RUNX3 affects preferentially the innervation of specific MNpools, we used Cholera toxin B subunit (CTB) injection into twomajor antagonistic groups of forelimb muscles: the biceps brachii

2

RESEARCH ARTICLE Development (2019) 146, dev181750. doi:10.1242/dev.181750

DEVELO

PM

ENT

(flexor muscles) and the triceps brachii (extensor muscles, Fig. 3A).Biceps and triceps CHAT+ MNs (MNsBiceps and MNsTriceps,respectively) were retrogradely traced by injecting CTB in theirrespective muscles at postnatal day (P) 1, and their proprioceptiveafferent terminals were examined at P2 by quantifying their numberof contacts with VGLUT1+ boutons (Pecho-Vrieseling et al., 2009;Surmeli et al., 2011). The majority of MNsBiceps were found

essentially in C5-6 spinal segments, while MNsTriceps were found inC7-T1 spinal segments, as previously shown (Vrieseling and Arber,2006). Gamma MNs, identified by their small size, were notconsidered for analysis. Quantification of the number of VGLUT1+

boutons making synapses with MNsBiceps revealed a 40% decreasein AdvCre;Runx3fl/fl mice compared with Runx3fl/fl controllittermates (Fig. 3B,C). Strikingly, the majority of VGLUT1+

Fig. 1. Loss of cell identity in subgroups of PSNs following conditional targeting of RUNX3 after peripheral innervation. (A) Scheme representing thesuccessive developmental steps of PSNs, which contribute to sensorimotor circuit. Early specification of PSNs (i) is followed by peripheral axonal growth andmuscle targeting (circa. E14) (ii). After peripheral innervation, central afferents of PSNs project to the intermediate and then ventral regions of the spinal cord tocontact interneurons andmotor neurons (iii). (B) Ablation ofRunx3 from sensory neurons usingAdvCremice. At E13.5, RUNX3 expression is detectable in TRKC+

neurons with tdTomato (RFP) starting to be expressed in few neurons, while at E15.5, the recombination is fully efficient, all neurons expressing tdTomato andRUNX3 are strongly reduced in number. Scale bar: 50 µm. (C) Quantification of B, showing the recombination efficiency in TRKC+/RUNX3+ in AdvCre;Runx3fl/fl

mice (n=3). ***P≤0.001; Student’s t-test. Data are mean±s.e.m. (D) Immunostaining for NF200 and RUNX1 on DRG sections from P0 AdvCre;Runx3fl/fl andRunx3fl/fl animals identifies all myelinated sensory neurons (mechanoreceptive and proprioceptive neurons) and a large majority of nociceptive neurons(Lallemend and Ernfors, 2012; Gascon et al., 2010). Scale bar: 50 µm. (E) Quantification of D reveals absence of cell death in DRG neurons in the conditionalRunx3 mutants at P0. P>0.05, Student’s t-test. Data are mean±s.e.m. (n=3). (F-K) Immunostaining for PSNs markers (F,H,J) and their quantification (G,I,K) inAdvCre;Runx3fl/fl and Runx3fl/fl P0 animals (n=3 per genotype). Scale bars: 100 µm in F; 50 µm in H,J. n=3 per genotype; *P≤0.05; Student’s t-test. (G) Data aremean±s.e.m.

3

RESEARCH ARTICLE Development (2019) 146, dev181750. doi:10.1242/dev.181750

DEVELO

PM

ENT

boutons synapsing ontoMNsTriceps were lost in the mutant (Fig. 3D,E),suggesting that RUNX3 expression in late embryonic stage has aselective role in the formation of sensorimotor connections betweenPSNs and defined pools of MNs.

Selective loss of the peripheral innervation of PSNs in theabsence of RUNX3We next wondered whether the preferential sensorimotorconnection deficits observed in AdvCre;Runx3fl/fl mice arereflected in the peripheral targeting of limb muscles by PSNs.MSs form after interaction between the sensory nerve endings of thePSNs and the targeted muscle fibers at ∼E14 (Tourtellotte et al.,2001). Here, we quantified at P0 the VGLUT1+ sensory endingsassociated with MSs as readout of muscle innervation by PSNs(Fig. 4A) (Poliak et al., 2016). The GTO-associated nerve endingsof the PSNs were not considered in the analysis. We focused on twopairs of antagonistic muscles spanning the proximo-distal axis ofthe forelimb: the extensors triceps and extensor carpi radialis (ECR);and the flexors biceps and flexor carpi radialis (FCR). Strikingly,

while the biceps and FCR muscles exhibited, respectively, a 30%decrease and normal incidence of VGLUT1+ PSN nerve endings inAdvCre;Runx3fl/fl mice, the triceps and ECR muscles displayed alarge reduction in PSN nerve endings (Fig. 4B; Fig. S3A,B). Inaddition, the few remaining MSs observed in triceps and ECRmuscles of AdvCre;Runx3fl/fl mutants were systematically of smallersize (Fig. 4A). These analyses at the forelimb level revealed aspecific role for RUNX3 in PSN specification that affects MSdifferentiation in some defined muscle groups. They also show thatthe loss of intraspinal PSN axons targeting specific pools of MNscorrelates well with the lack of sensory nerve endings in the limbmuscles they innervate.

The observation that this innervation deficit is not associated withsensory neuron cell death in the AdvCre;Runx3fl/fl mutants at P0raises the issue of what happens to those particular PSNs that havelost contact with their peripheral target. The peripheral projectionsof brachial PSNs make dorso-ventral choice within the forelimbbefore E12.5 (Kania et al., 2000; Huettl et al., 2011), and reach theirperipheral target at E14 (Tourtellotte et al., 2001). By deleting

Fig. 2. Central afferentation deficit of PSNs after conditional targeting of RUNX3. (A) Representative regions of the spinal cord analyzed. To reveal centralafferent terminations of PSNs, we used VGLUT1 immunostaining on cross-sections of spinal cord. The pattern of VGLUT1 reactivity was analyzed in threereference regions: the intermediate zone (IZ), the ventromedial (M) and the ventrolateral (L) regions. Scale bar: 100 µm. (B) VGLUT1 expression in AdvCre;Runx3fl/fl and Runx3fl/fl DRG sections at P0. Scale bar: 50 µm. (C) Quantification of the number of VGLUT1+ neurons per DRG section (left panel) and VGLUT1intensity per cell (right panel) (from data in B) reveals absence of change in VGLUT1 expression in DRG from AdvCre;Runx3fl/fl mice (n=3). (D,E) Centralinnervation of PSNs inAdvCre;Runx3fl/fl andRunx3fl/flmice at C5 (D) andC8 (E), as revealed by VGLUT1 immunostaining. Scale bar: 100 µm. (F) Quantification ofthe density of VGLUT1 staining in D and E in regions defined in A on one side of the spinal cord reveals deficits in central ingrowth of PSN afferents in conditionalRunx3 mutant mice (n=4 per genotype). A greater difference was observed in the lateral and medial (L and M) regions of the ventral spinal cord, whichcorresponds to the innervation of the MN pools (CHAT+ in A). **P≤0.01, ***P≤0.001; Student’s t-test. Data are mean±s.e.m. (n=3). (G) Immunostaining forperipherin (PERI) on spinal cord sections shows complete absence of central PSN afferents in Runx3−/−;Bax−/− compared with AdvCre;Runx3fl/fl mice (see redarrowheads). Scale bar: 100 µm. (H) Similar to G, at P0 VGLUT1 immunostaining on spinal cord sections confirms the absence of central axon growth of PSNs inRunx3−/−;Bax−/− mice (see red arrowheads), a phenotype that differs from AdvCre;Runx3fl/fl mice (see D,E). Scale bar: 100 µm.

4

RESEARCH ARTICLE Development (2019) 146, dev181750. doi:10.1242/dev.181750

DEVELO

PM

ENT

RUNX3 after this period using the AdvCre driver, it is thus unlikelythat the PSNs losing their synaptic connection and identity willretract their projections to target another opposite muscle ordifferent sensory organ such as the skin. Indeed, both the intrinsicsignaling and environmental cues might not enable these cells toadopt a different sensory cell fate or trajectory. However, althoughfuture studies will be necessary to resolve this issue, the observedloss of identity features could eventually lead to cell death at laterstage, as shown after the loss of many terminal selector genes thatare needed to maintain a differentiated state (Deneris andHobert, 2014).

Role of muscle target and NT3 signaling in sustained RUNX3expressionWe next asked whether the phenotypic defects observed in PSNs inour AdvCre;Runx3fl/flmutants could reflect a role for muscle-derivedsignals in PSN afferent specification. To assess this, we analyzedRUNX3 expression in Lbx1 null mice, in which all extensor musclesare lacking at forelimb level although a large number of flexormuscles still develop (Gross et al., 2000; Brohmann et al., 2000).However, this phenotype is not accompanied by a loss in PSNnumber or by a defect in the peripheral outgrowth of PSN afferentswithin the limbs (Poliak et al., 2016). Examining brachial DRGneurons from E18.5 Lbx1−/− mice, we observed a 40% reductionin the number of neurons expressing RUNX3 (Fig. 5A,B).

Interestingly, the remaining RUNX3+ cells in Lbx1 mutantsexhibited reduced levels of expression similar to low RUNX3-expressing cells in wild-type animals [Lbx1+/+, 1±0.077; Lbx1−/−,0.401±0.069 (expressed as mean±s.d.); P<0.01, normalized toLbx1+/+], indicating that RUNX3 levels in PSNs might correlatewith the specific target muscle groups they innervate. To test this,we analyzed backfilled PSNs following retrograde tracing usingrhodamine dextran (Rh.dex.) injection in triceps and biceps fromE16.5 wild-type embryos (Fig. 5C). Biceps- and triceps-innervatingPSNs were primarily found in C5 and C8 segments, and expressed(on average) low and high levels of RUNX3, respectively, whencompared with the whole PSN population (Fig. 5C-E; Fig. S4A).Together, these data suggest that RUNX3 expression and its level inPSNs depend on limb muscles at late embryonic stages. To supportthis, we analyzed Bax−/− mice devoid of NT3 signaling and inwhich PSNs completely fail to project into limb muscles but surviveeven in reduced neurotrophic factor conditions (Patel et al., 2003).RUNX3 expression was dramatically decreased in E15.5 TrkC−/−;Bax−/− DRG and virtually absent at birth (P0; Fig. 5F,G; Fig. S4B).Its expression in E13.5 TrkC−/−;Bax−/− DRG, i.e. prior to targetinnervation, however, was unchanged (Fig. S4B), indicating a latefunction of muscle-derived signals, following sensory afferent-muscle fiber contact, in the regulation of RUNX3 expression inPSNs. NT3 itself was shown to be a plausible candidate forcontrolling later aspects of PSNs specification (Wang et al., 2007;

Fig. 3. Muscle-selective differential penetrance of central deficits of PSN connectivity in RUNX3 conditional mutants. (A) Experimental schemedescribing muscle injection of the antagonistic muscles biceps and triceps, and retrograde labeling of specific MN pools at spinal segments C5-C6 for the bicepsor segments C7-T1 for the triceps. (B) Immunostaining of spinal cord cross-sections representing MNs (CHAT+, in blue) traced by the CTB (red) from initialinjection in the biceps. PSN synaptic contacts with the MNs are visualized by the VGLUT1+ synaptic bouton (green). Scale bars: 10 µm. (C) Quantification of 38MNs (from B) in Runx3fl/fl and 39 MNs in AdvCre;Runx3fl/fl. ***P≤0.001; Student’s t-test. Data are mean±s.e.m. (n=3). (D) Immunostaining of spinal cord cross-sections representing MNs (CHAT+, in blue) traced by the CTB (red) from initial injection in the triceps. PSN synaptic contacts with the MNs are visualized by theVGLUT1+ synaptic bouton (green). Scale bars: 10 µm. (E) Quantification of a total of 44 MNs (from D) in Runx3fl/fl and 46 MNs in AdvCre;Runx3fl/fl. ***P≤0.001;Student’s t-test. Data are mean±s.e.m. (n=3).

Fig. 4. Muscle-specific MSs deficits in conditionalRunx3 mutants at birth. (A) Immunostaining forVGLUT1 (for MSs) and myosin on cross-sections frombiceps and triceps. (B) Quantification reveals a muscle-selective MS deficiency in AdvCre;Runx3fl/fl with asignificant decrease in the numbers of MSs in triceps(n=3 animals). **P≤0.01, ****P≤0.0001; Student’s t-test.Data are mean±s.e.m. Scale bar: 100 µm.

5

RESEARCH ARTICLE Development (2019) 146, dev181750. doi:10.1242/dev.181750

DEVELO

PM

ENT

De Nooij et al., 2013). Notably, elevated levels of NT3 in muscleslate during embryonic development has been shown to disrupt theselective pattern of synaptic connections between sensory afferentsand MNs (Wang et al., 2007). In addition, various levels of NT3 inhindlimb muscles have been correlated with subclass-specificsensitivity of PSNs to cell death following loss of ER81 (De Nooijet al., 2013). This heterogeneity in NT3 levels among muscles wasalso observed here in forelimb muscles using β-gal activity in

whole-mounted Ntf3LacZ/+ E15.5 embryos and confirmed usingqPCR (Fig. 5H,I). To directly test the role of NT3 on RUNX3expression at late embryonic stage, we cultured E15.5 DRG explantsfrom Bax−/− embryos for 2 days in the presence or absence of NT3(Fig. 5J). Sensory neurons from Bax−/− DRG could survive severaldays in vitro without neurotrophin signaling (Lentz et al., 1999;Zhong et al., 2007). We found higher levels of RUNX3 expressionin the presence of NT3 compared with the control condition

Fig. 5. RUNX3 dependence of PSN connectivity correlates with RUNX3 levels and muscle target NT3 levels. (A) Immunostaining for ISL1 and RUNX3 onDRG sections from Lbx1−/− and Lbx1+/+mice. Scale bar: 50 µm. (B) Quantification of RUNX3+ neurons (fromA) in Lbx1−/− (n=2) and their control littermates (n=4)at brachial levels shows significant reduction in the full mutants compared with control animals. **P≤0.01. Data are mean±s.e.m. (C) Experimental design. (Left)Rhodamine dextran (Rh. dex.) injection in specific muscle will retrogradely trace their innervating PSNs in the DRG from E16.5 wild-type animals. The dextran isinjected in biceps in one of the forelimbs and in triceps in the contralateral limb. (Right) Distribution within DRG of the PSNs innervating biceps and triceps,showing biceps- and triceps-innervating PSNs located mostly in DRG C5 and C8, respectively (data from 5 animals for triceps and 11 animals for biceps). (D,E)Quantification of RUNX3 expression per cell in retrogradely traced PSNs (as in C) versus all RUNX3+ PSNs after injection of Rh. dex. in triceps (analysis at C8level, D) or in biceps (analysis at C5 level, E). **P≤0.01; Student’s t-test. Data are mean±s.e.m. (single neurons analyzed from five embryos). (F) RUNX3expression is largely reduced in TrkC−/−;Bax−/− P0 mice: a mouse model of peripheral outgrowth deficits. Scale bar: 100 µm. (G) Quantification of F reveals analmost complete absence of RUNX3 expression in TrkC−/−;Bax−/− P0 mice compared with their control littermates (n=2). **P≤0.01; Student’s t-test. Data aremean±s.e.m. (H) X-Gal reaction on E15.5 in Nft3Lacz/+ forelimb embryos show heterogeneous muscle-specific expression of NT3. There is a large difference inNT3 levels between biceps and triceps. (I) Quantification of NT3 mRNA (Nft3) in the biceps (Bic) and triceps (Tri) shows a twofold increase in triceps comparedwith biceps in control animals (Ctr). P=0.052; Student’s t-test (n=2 samples from four animals). Data are mean±s.e.m. (J) Bax−/− mice DRG explants (E15.5) inculturewith or without NT3 for 48 h in vitro (HIV) reveal a decreased expression of RUNX3 in the absence of NT3. Scale bar: 100 µm. (K) Quantification of J showsa significant increase of RUNX3 intensity per cell, while the number of positive neurons remains unchanged. **P≤0.01; Student’s t-test. Data are mean±s.e.m.(n=3). (L) Immunostaining for ISL1, RET and RUNX3 on spinal cord (SC) sections from Hb9Cre;Islet2DTA mice shows complete absence of MNs at E16.5.Scale bar: 50 µm. (M) Immunostaining for TRKC, RUNX3 and ISL1 on DRG sections fromHb9Cre;Islet2DTA and Islet2DTAE16.5 mice shows no deficits in RUNX3and TRKC expression in the absence of MNs, confirmed by the quantification of the number of PSNs in Hb9Cre;Islet2DTA and Islet2DTA (n=2, right panel).Scale bar: 20 µm.

6

RESEARCH ARTICLE Development (2019) 146, dev181750. doi:10.1242/dev.181750

DEVELO

PM

ENT

(Fig. 5K). Supporting these data, addition of NT3 to DRG explantsfrom HH27 chicken embryos resulted in marked upregulation ofRunx3 mRNA levels compared with untreated DRG explants(Fig. S4C,D). Thus, after peripheral innervation, distinct levels ofRUNX3 expression in PSN subgroups are controlled by muscle-derived signals and involve NT3 signaling.We also considered whether MNs, through direct axon-axon

contact or indirect signaling, could act on RUNX3 expression inPSNs. To test this, we usedHb9cre;Isl2DTAmice, in which diphtheriatoxin (DTA) is selectively expressed in HB9+/ISL2+ MNs, ablatingMNs as they exit the cell cycle, before peripheral innervation (Yanget al., 2001) (Fig. 5L). In the absence of MNs, PSNs density in DRGis unchanged, and their nerve endings can still be found inassociation with MSs (Poliak et al., 2016). In Hb9cre; Isl2DTA miceexamined at E16.5, RUNX3 expression in PSNs was foundunchanged compared with Isl2DTA mice (Fig. 5M). These datathus argue against the idea that MNs could provide signalsnecessary for regulating or maintaining RUNX3 in PSNs at lateembryonic stages.

DISCUSSIONMuscle sensory feedback is essential for controlling coordination oflocomotor behavior. It is conveyed by different subgroups of PSNsthat form specific connections with second order neurons and MNsin the spinal cord. Despite the importance of this feedback in motorbehavior, our knowledge of the molecular programs that control theassembly of the sensorimotor circuits remains limited. This isessentially due to the fact that the major genetic determinants ofPSNs identity, which continue to be expressed in the matureneurons, are also necessary for their early peripheral outgrowth andsurvival, limiting the study of their direct function in stage-specificmaintenance of PSNs identity and connectivity. Here, we haveused conditional Cre/loxP gene-targeting approach to delete themajor regulatory factor Runx3 in PSNs after the period of naturalcell death and peripheral innervation. We find that RUNX3 iscontinuously required throughout development to maintainspecific identity features of particular subgroups of PSNsinnervating the limb, independent of cell death. We also showthat the maintained expression of RUNX3 in PSNs is regulated bymuscle target-derived signals and NT3 signaling which isdifferentially required in subgroups of PSNs to setup specificsensorimotor connections.

Dedicated maintenance factors for PSN subgroupsThe generation of proprioceptive afferents during embryonicdevelopment follows a systematic sequence of events, fromneurogenesis, early cell fate specification and target innervation tofiner aspects of PSN subgroup-specific identity and the formation ofthe sensorimotor circuits (Lallemend and Ernfors, 2012). Hence,soon after neurogenesis, TRKC+ PSNs acquire RUNX3 expression,which is necessary to consolidate a proprioceptive fate and direct theextent of peripheral outgrowth of PSNs projections (Lallemendet al., 2012; Levanon et al., 2002; Kramer et al., 2006). As theiraxons grow peripherally, PSNs, through target-derived NT3signaling, will initiate expression of ER81, which later will definethe survival competence of particular subgroups of PSNs incorrelation with their target muscles (Patel et al., 2003; De Nooijet al., 2013). At the same time, different subtypes of PSNs emergeon the basis of their preferential expression of recognition moleculesdefined by interactions with signals notably from the limb (Poliaket al., 2016; Wu et al., 2019). In this context, it has been unclearwhether early specification markers continue to be crucial during

late embryogenesis for preservation and maturation of the sensoryneuron type identity and connectivity. Our data indicate thatRUNX3 is not essential for the survival of PSNs followingperipheral innervation. This suggests that the loss of PSNs observedin Runx3-null mice at E12.5 is likely a consequence of the severereduction in TRKC expression and survival signaling within PSNsduring the developmental cell death period (Lallemend et al., 2012;Levanon et al., 2002; Kramer et al., 2006). This is reminiscent of arecent study exploring the role of BRN3A in the diversification ofsensory neurons of the cochlea, where conditional deletion ofPou4f1 (encoding BRN3A) from E14.5 (while it is first expressed atE10.5) did not affect the survival of cochlear neurons as seen in thefull mutant where TRKC is found to be downregulated (Huanget al., 2001; Sherrill et al., 2019). In contrast, classic key molecularfeatures of PSN identity, such as ER81, TRKC and PV expression,showed a deficit in our conditional Runx3 mutant. Interestingly,only a subgroup of PSNs, representing about half of its population atthe forelimb level, showed a loss of identity marker expression. Thisimplies the existence of two distinct subgroups of PSNs at limblevels based on their dependence on RUNX3 expression for the lateraspect of their differentiation. In support of this, key molecularmarkers of PSNs are found unchanged in thoracic DRG in ourconditional Runx3mutant mice. One possible mechanism that couldparticipate in the development of these two distinct subgroups ofPSNs is the graded level of RUNX3 activity itself. A previous studyin chicken embryos using in ovo electroporation suggested that thestatus of RUNX3 expression could participate in the segregation ofthe different subtypes of PSNs (innervating GTOs or MSs), notablyby regulating their central projection trajectory (Chen et al., 2006).Our data in mice indicate instead that the levels of RUNX3expression and its role in PSN diversification might correlate withthe identity of the target muscles being innervated, with meanRUNX3 levels decreasing in remaining PSNs projecting to forelimbflexor muscles in Lbx1 null mice, which are devoid of extensormuscles (Gross et al., 2000; Brohmann et al., 2000). Thesecontrasting data could indicate a species difference. In support,siRNA-based extinction of RUNX3 expression in embryonicchicken PSNs did not lead to loss of TRKC (Chen et al., 2006),which would be expected from the data found in mice (Lallemendet al., 2012; Kramer et al., 2006). Our data remind us, however, ofthe differential activity of the NT3-ER81 signaling pathway in PSNsand its role in delineating two classes of PSNs based on theirdependence on ER81 for their survival (De Nooij et al., 2013).Although results from our work would need further investigation ofa direct link between RUNX3 levels and PSN diversification,graded levels of transcription factor activity may play a crucial rolein the development of neuronal identity. In this context, the regionaldifferences in the expression at the target of signaling cues mightserve at the nerve endings as key factors eliciting various strengthsof retrograde signaling operating in the neurons to drive theiridentity. A very good example is the development of the trigeminalganglion, in which the coincident expression of both BDNF andBMP4 in discrete regions of the peripheral target field plays animportant role in controlling the development of specific neuronalpools within the ganglion (Ji and Jaffrey, 2012). Distinct signalsderived from limb mesenchyme have also been hypothesized todrive the subclass identity of PSNs. In a similar way, we show herethat the presence of muscles is necessary for maintaining theexpression of RUNX3 in PSNs. Moreover, NT3, which is expressedat various levels in developing skeletal muscles (Fariñas et al., 1996;De Nooij et al., 2013), can directly regulate RUNX3 expression exvivo. Hence, it is conceivable that various extents of NT3-TRKC

7

RESEARCH ARTICLE Development (2019) 146, dev181750. doi:10.1242/dev.181750

DEVELO

PM

ENT

signaling activities between PSNs subgroups innervating distinctmuscles might regulate their later cell identity aspects throughgraded RUNX3 expression. Following peripheral innervation, suchdifferential activities among distinct PSNs subgroups couldeventually serve to link the targeted muscles with the centralneurons in the spinal cord the PSNs must connect.

Selectivity of sensorimotor connectivityCertainly the most studied circuit in sensorimotor networkconnecting PSNs and spinal cord neurons is the reflex arc inwhich group Ia PSN afferents make specifically strong connectionswith MNs supplying the same muscle and weaker connections withmotor neurons supplying synergistic muscles, while no connectionis established withMNs of antagonistic muscles (Eccles et al., 1957;Frank and Mendelson, 1990). During the formation of this circuit,an elegant study has shown that the topography of the PSN endingsalong the proximodistal axis of the limb could predict thedorsoventral location and identity of their target MNs, suggestingan area-specific targeting of PSNs within the ventral spinal cord(Surmeli et al., 2011). Others emphasized the importance of a directMN-dependent mechanism that involves cell-to-cell recognitionwith PSNs and the use of Sema3E-plexin D1 signaling for theformation of sensorimotor connectivity patterns (Fukuhara et al.,2013; Pecho-Vrieseling et al., 2009). This implies a stepwisetargeting approach making use of different signaling and strategiesthat PSN must use to eventually synapse to the appropriate MNs.Our data here on the selective function of RUNX3 in the spinal cordingrowth of particular subgroups of PSNs indicate another level ofregulation by which only a subset of neurons critically depends onthe activity of a specific transcriptional program to project within thespinal cord and eventually develop a proper connectivity.Interestingly, this RUNX3 dependence of PSN-MN connectionsexhibits a preference for neurons innervating forelimb extensors,when compared with flexors and axial muscles at limb level, as wellas thoracic muscles. In adult animals, such a deficit of musclespindles in limb extensors, in acute situation, would most likelycause a substantial decrease in muscle tone of extensors, resulting ina change in the basic body configuration (a configuration with moreflexed limbs as compared with control) and a decrease in efficacy offeedback postural corrections. In our mutant mice, due to adaptiveplastic changes that could take place during early life, these deficitscould be compensated for, e.g. by an increase in the activity ofneurons of the vestibulospinal tract leading to an increase inextensor muscle tone and/or by a substitution of proprioceptiveinformation (which in control subjects plays a crucial role in thegeneration of postural corrections) by visual and vestibular signals.However, the most likely specific aspects of postural correctionsmight still differ from those observed in control animals.The RUNX3-dependent bias to neurons innervating forelimb

extensors, compared with flexors muscles, evokes the ER81dependence of PSNs for their survival, where PSNs innervatingthoraco-hypaxial and axial muscles exhibit an almost completedependence on ER81, whereas those innervating hindlimb musclesexhibit a mosaic, muscle-by-muscle, sensitivity to ER81inactivation (De Nooij et al., 2013). Strikingly, the levels of NT3expression in individual muscles strongly correlated with thisdependence on ER81 of the survival of particular PSNs subgroups.Although our study only considers selected sensorimotorconnections, it suggests a role for peripheral gradation of NT3 insetting up the sensitivity of PSNs for RUNX3-dependentsensorimotor connectivity, as described above. Even if furtherprograms are certainly required to create the fine topography of

PSNs to MNs connectivity, this earlier step already distinguishespopulations of neurons as they need to enter the spinal cord, asshown in chick and mouse for the selectivity of the sensorimotorcircuits by regionally restricted limb signals (Wenner and Frank,1995; Poliak et al., 2016). This could serve to coordinate spatiallyand temporally separated developmental events, as observed duringthe development of olfactory receptor neurons and their selectiveconnectivity with their glomerular target (Chou et al., 2010). Theimplication of these results is also that, within proprioceptors,higher levels of RUNX3 might engage a distinct set oftranscriptional targets that are necessary for their propermaturation. The lower RUNX3-expressing PSNs would thenrequire the activation of another set of transcriptional regulators.Such a mechanism, which most likely involves differences in target-derived signals, including neurotrophins, would thus participate in aspatial patterning of transcription factor activities among sensoryneurons.

MATERIAL AND METHODSAnimalsWild-type C57BL6 mice were used unless specified otherwise. Runx3−/−,AdvillinCre (AdvCre) Lbx1−/−, Hb9Cre and Isl2DTA mouse strains have beendescribed elsewhere (Levanon et al., 2002; Yang et al., 2001; Gross et al.,2000; Zhou et al., 2010). Runx3tm1ltan (Runx3fl), Bax−/−, TrkC−/− and Ai14mice were purchased from Jackson Laboratories, and NT3−/− fromMMRRC. Both male and female animals were included in this study,except for the AdvCre; Runx3fl/fl, where only males were analyzed because ofweak expression of Cre in oocytes and possible leakage or generation of‘deletor’ mice. Animals were group-housed, with food and water availablead libitum, under 12 h light-dark cycle conditions. Fertile white Leghorneggs were incubated at 38°C and embryos were staged according toHamburger-Hamilton (HH) tables. All animal work was performed inaccordance with the national guidelines and approved by the local ethicscommittee of Stockholm, Stockholms Norra djurförsöksetiska nämnd.

In vitro cultures of whole DRGWhole DRG were cultured on coated plates (5% matrigel in ice-cold PBS)with N2 medium (DMEM-F12/glutamax medium with N2 supplement;Gibco) supplemented with pen/strep, gentamicin, the pan-caspase inhibitorQ-VD-Oph (2 μM, Sigma) and with NT3 (Peprotech), retinoic acid, bFGFand IGF when specified, as previously described (Hadjab et al., 2013). DRGanalyzed for mRNA expression were cultured directly in eppendorf tubes.

qPCRTissue was freshly dissected and directly placed into lysis buffer. RNAwasextracted using the Qiagen RNAeasy Mini Kit according to themanufacturer’s instructions, including DNase I to degrade potentialgenomic DNA contamination. The amount of RNA was quantified usingthe Qubit RNA BR Assay Kit, and 50 ng of each biological replicate usedfor reverse transcription using Biorad iScript in a 20 µl reaction according tothe manufacturer’s instructions. Produced cDNAs were used to analyzetranscript levels using real-time PCR in a Biorad CFX96 Real-time Systemand using Biorad iTaq Universal SYBR Green Supermix in 20 µl reactionsfor 40 cycles (95°C for 10 s, 60°C for 20 s, 72°C for 30 s followed byfluorescence measurement at 75°C to reduce the influence of primer dimeron quantification). Primers forNft3 and Runx3were designed using Primer3Blast (NCBI). Primers for Nft3, Ntrk3 (TRKC), Isl1 and Runx3 weredesigned using Primer3 Blast (NCBI). Gapdh was used as a reference gene.A full list of primers used is given below. Negative controls did not amplify.Data were analyzed using Bio-Rad CFX manager. Gapdh, AACTCCCACT-CTTCCACCTTC and GATAGGGCCTCTCTTGCTCAG; Nft3, ATAAAA-TTCGTGTGCTTGCCTTCC and GAGAGCCCAATCACAAAACAAGG;Runx3, ACCAAGTGGCGAGATTTAACGA and ACGGTGACTTTAAT-GGCTCGG; Isl1, AAAAGAAGCATTATGATGAAGCAA and CATGTC-TCTCCGGACTAGCAG; Ntrk3, TGATCCTCGTGGATGGGCAG andCTTCACCAGCAGGTTGGCTCC.

8

RESEARCH ARTICLE Development (2019) 146, dev181750. doi:10.1242/dev.181750

DEVELO

PM

ENT

ImmunostainingMouse embryos were collected and fixed for 1 to 6 h at +4°C (4% PFA inPBS) depending on the stage, washed in PBS three times (30 min each),equilibrated in 20% and 30% sucrose in PBS, embedded in OCT (Tissue-Tek) and cryosectioned at 14 μm. Sections were incubated for one or twonights with primary antibodies diluted in blocking solution (2% donkeyserum, 0.0125% NaN3, 0.5% Triton X-100 in PBS). Primary antibodiesused were: rabbit anti-RUNX3 (a gift from T. M. Jessell and D. Levanon,Columbia University and the Weizmann Institute of Science), goat anti-TRKA (R&D Systems, AF1056, 1/500), goat anti-TRKB (R&D Systems,AF1494, 1/500), goat anti-TRKC (R&D Systems, AF1404, 1/500), goatanti-RET (R&D Systems, AF482, 1/500), mouse anti-ISL1 (DevelopmentalStudies Hybridoma Bank, 39.4D5, 1/200), rabbit anti-VGLUT1 (SYSY135303, 1/1000), chicken anti-RFP (Rockland, 600-401-379, 1/500), goatanti-ChAT (Millipore, AB144p, 1/1000), mouse anti-NF200 (Sigma,N0142), rabbit anti-peripherin (Millipore, AB1530, 1/500), rabbitanti-PEA3 (a gift from S. Arber, Friedrich Miescher Institute, Basel,Switzerland) and mouse anti-myosin (DSHB, F59-s, 1/50). After washingwith PBS, Alexa Fluor secondary antibodies (Life Technology; 1:500 inblocking solution) were applied overnight (at +4°C). Samples were thenwashed in PBS and mounted in DAKO fluorescent mounting medium.Staining was imaged by confocal microscopy (Zeiss LSM700 or LSM800)using identical settings between control and experimental groups. Opticalpinholes were 2 μm under 20× magnification unless specified otherwise.

Quantification of neuronsFor cell type counting quantifications, ImageJ software was used. Onlyneurons with a visible nucleus were considered for analysis. Quantification ofmolecular markers in the DRG was carried out on five DRG sections/animal,selected from the most equatorial region of each DRG and covering thesegments C5-T1 (see figure legends for n values and genotypes).

Quantification of muscle spindlesLimb muscle VGLUT1+ muscle spindles were counted on longitudinalsections of whole muscles.

Quantification of PSN collateral densityQuantitative analysis of PSN axon terminals in the ventrolateral horn of thespinal cord was performed using ImageJ analysis software. For eachsegment level (C5 and C8) the total VGLUT1+ collateral surface area(expressed as squared microns) was measured within a confinedintermediate zone, ventromedial area and ventrolateral area of the ventralhorn (adjusted for each segmental level). The surface areas analyzed were asfollows: C5 ventrolateral, 210,000 µm2; C5 ventromedial, 112,000 µm2; C5intermediate zone, 37,400 µm2; C8 ventrolateral, 160,000 µm2; C8ventromedial, 63,000 µm2; and C8 intermediate zone, 30,000 µm2. Foreach genotype, measurements for all segmental levels were performed fromat least two individual experiments and three z-stacks per segment.

Quantification of sensory synaptic contacts with motor neuronsQuantification of VGLUT1+ sensory bouton contacts between PSNs andmotor neurons somata was performed using 0.5 µm confocal Z-scans of 10to 20 µm thick sections. Motor neuron surface area was determined usingImage J. Synapses were counted manually and filtered for sizes largerthan 1 µm2.

β-Galactosidase stainingLacZ expression was detected by staining using X-Gal (5-bromo-4-chloro-3-indolyl β-d-galactoside) for β-galactosidase activity as describedpreviously (Lallemend et al., 2012).

Retrograde labeling of motor and sensory neuronsNewborn mice were anesthetized on ice by hypothermia and muscles ofinterest were injected with 1% cholera toxin B subunit/Alexa555 (CTB555).The amount of CTB555 was enough to fill the whole muscle. The followingday, spinal columns were dissected and fixed in 4% PFA for 6 h thenprocessed for immunostaining and analysis. Sensory neuron tracing was

performed ex vivo by injection of fluorescently labeled dextrans (3000MW,Invitrogen) in limb muscles using tightly fitting glass capillaries. Followinginjection, the ex vivo preparation (brachial region with associated limbs) wasincubated for 8 h in oxygenated ACSF before fixation.

StatisticsData were analyzed using GraphPad Prism 6 and are expressed as mean±s.e.m. The statistical test performed is reported in the figure legend. t-testswere two-sided (*P≤0.05, **P≤0.01, ***P≤0.001). No animals or datapoints were excluded from the analyses. No statistical methods were used topre-determine sample size but our sample sizes are similar to those generallyemployed in the field.

AcknowledgementsWe thank Prof. Yoram Groner and Ditsa Levanon for the Runx3−/− mouse line. Wethank the CLICK imaging Facility supported by the Knut and Alice WallenbergFoundation.

Competing interestsThe authors declare no competing or financial interests.

Author contributionsConceptualization: S.H., F.L.; Methodology: Y.W., H.W., P.F., S.W., C.P., R.-E.H.,A.H.B., Y.X.-F.; Validation: H.W., P.F., S.W.; Formal analysis: Y.W., H.W., P.Z., P.F.,S.W., C.P., C.B., T.G.D.; Investigation: S.H., F.L.; Resources: P.Z., G.C., C.B., S.T.,O.K., P.E., T.G.D.; Data curation: S.H., F.L.; Writing - original draft: S.H., F.L.; Writing- review & editing: Y.W., H.W., P.F., S.W., C.P., G.C., C.B., Y.X.-F.; Visualization:S.H., Y.W.; Supervision: S.H., F.L.; Project administration: S.H., F.L.; Fundingacquisition: S.H., F.L.

FundingThis work was supported by StratNeuro (2013-0237), the Vetenskapsrådet (2012-04708), a Karolinska Institute doctoral grant the Knut och AliceWallenbergs Stiftelse(Wallenberg Academy Fellow), the Hjarnfonden (FO2014-0048), the KarolinskaInstitutet (Faculty Funded Career Position) and the Ragnar Soderbergs stiftelse(M48/12) (Ragnar Soderberg Fellow in Medicine), and by a Ming Wai Lau researchgrant. F.L. is a Ragnar Soderberg fellow in Medicine, a Wallenberg Academy Fellowin Medicine and a Ming Wai Lau Center investigator. Deposited in PMC forimmediate release.

Supplementary informationSupplementary information available online athttp://dev.biologists.org/lookup/doi/10.1242/dev.181750.supplemental

ReferencesArber, S., Ladle, D. R., Lin, J. H., Frank, E. and Jessell, T. M. (2000). ETS gene

Er81 controls the formation of functional connections between group Ia sensoryafferents and motor neurons. Cell 101, 485-498. doi:10.1016/S0092-8674(00)80859-4

Brohmann, H., Jagla, K. and Birchmeier, C. (2000). The role of Lbx1 in migrationof muscle precursor cells. Development 127, 437-445.

Chen, A. I., de Nooij, J. C. and Jessell, T. M. (2006). Graded activity of transcriptionfactor Runx3 specifies the laminar termination pattern of sensory axons in thedeveloping spinal cord. Neuron 49, 395-408. doi:10.1016/j.neuron.2005.12.028

Chou, Y.-H., Zheng, X., Beachy, P. A. and Luo, L. (2010). Patterning axontargeting of olfactory receptor neurons by coupled hedgehog signaling at twodistinct steps. Cell 142, 954-966. doi:10.1016/j.cell.2010.08.015

Dasen, J. S. (2009). Transcriptional networks in the early development of sensory-motor circuits. Curr. Top. Dev. Biol. 87, 119-148. doi:10.1016/S0070-2153(09)01204-6

Dasen, J. S., Tice, B. C., Brenner-Morton, S. and Jessell, T. M. (2005). A Hoxregulatory network establishes motor neuron pool identity and target-muscleconnectivity. Cell 123, 477-491. doi:10.1016/j.cell.2005.09.009

Deckwerth, T. L., Elliott, J. L., Knudson, C. M., Johnson, E. M., Jr, Snider, W. D.and Korsmeyer, S. J. (1996). BAX is required for neuronal death after trophicfactor deprivation and during development. Neuron 17, 401-411. doi:10.1016/S0896-6273(00)80173-7

De Marco Garcia, N. V. and Jessell, T. M. (2008). Early motor neuron pool identityand muscle nerve trajectory defined by postmitotic restrictions in Nkx6.1 activity.Neuron 57, 217-231. doi:10.1016/j.neuron.2007.11.033

Deneris, E. S. and Hobert, O. (2014). Maintenance of postmitotic neuronal cellidentity. Nat. Neurosci. 17, 899-907. doi:10.1038/nn.3731

De Nooij, J. C., Doobar, S. and Jessell, T. M. (2013). Etv1 inactivation revealsproprioceptor subclasses that reflect the level of NT3 expression in muscletargets. Neuron 77, 1055-1068. doi:10.1016/j.neuron.2013.01.015

9

RESEARCH ARTICLE Development (2019) 146, dev181750. doi:10.1242/dev.181750

DEVELO

PM

ENT

Eccles, J. C., Eccles, R. M. and Lundberg, A. (1957). The convergence ofmonosynaptic excitatory afferents on to many different species of alphamotoneurones. J. Physiol. 137, 22-50. doi:10.1113/jphysiol.1957.sp005794

Farin as, I., Yoshida, C. K., Backus, C. and Reichardt, L. F. (1996). Lack ofneurotrophin-3 results in death of spinal sensory neurons and prematuredifferentiation of their precursors. Neuron 17, 1065-1078. doi:10.1016/S0896-6273(00)80240-8

Farin as, I., Wilkinson, G. A., Backus, C., Reichardt, L. F. and Patapoutian, A.(1998). Characterization of neurotrophin and Trk receptor functions in developingsensory ganglia: direct NT-3 activation of TrkB neurons in vivo. Neuron 21,325-334. doi:10.1016/S0896-6273(00)80542-5

Frank, E. (1990). The formation of specific synaptic connections between musclesensory and motor neurons in the absence of coordinated patterns of muscleactivity. J. Neurosci. 10, 2250-2260. doi:10.1523/JNEUROSCI.10-07-02250.1990

Frank, E. and Mendelson, B. (1990). Specification of synaptic connectionsmediating the simple stretch reflex. J. Exp. Biol. 153, 71-84.

Fukuhara, K., Imai, F., Ladle, D. R., Katayama, K., Leslie, J. R., Arber, S., Jessell,T. M. and Yoshida, Y. (2013). Specificity of monosynaptic sensory-motorconnections imposed by repellent Sema3E-PlexinD1 signaling. Cell Rep. 5,748-758. doi:10.1016/j.celrep.2013.10.005

Gascon, E., Gaillard, S., Malapert, P., Liu, Y., Rodat-Despoix, L., Samokhvalov,I. M., Delmas, P., Helmbacher, F., Maina, F. andMoqrich, A. (2010). Hepatocytegrowth factor-Met signaling is required for Runx1 extinction and peptidergicdifferentiation in primary nociceptive neurons. J. Neurosci. 30, 12414-12423.doi:10.1523/JNEUROSCI.3135-10.2010

Gross, M. K., Moran-Rivard, L., Velasquez, T., Nakatsu, M. N., Jagla, K. andGoulding, M. (2000). Lbx1 is required for muscle precursor migration along alateral pathway into the limb. Development 127, 413-424.

Hadjab, S., Franck, M. C. M., Wang, Y., Sterzenbach, U., Sharma, A., Ernfors, P.and Lallemend, F. (2013). A local source of FGF initiates development of theunmyelinated lineage of sensory neurons. J. Neurosci. 33, 17656-17666. doi:10.1523/JNEUROSCI.1090-13.2013

Honig, M. G. (1982). The development of sensory projection patterns in embryonicchick hind limb. J. Physiol. 330, 175-202. doi:10.1113/jphysiol.1982.sp014336

Honig, M. G., Frase, P. A. and Camilli, S. J. (1998). The spatial relationshipsamong cutaneous, muscle sensory andmotoneuron axons during development ofthe chick hindlimb. Development 125, 995-1004.

Huang, E. J., Liu, W., Fritzsch, B., Bianchi, L. M., Reichardt, L. F. and Xiang, M.(2001). Brn3a is a transcriptional regulator of soma size, target field innervationand axon pathfinding of inner ear sensory neurons.Development 128, 2421-2432.

Huettl, R.-E., Soellner, H., Bianchi, E., Novitch, B. G. and Huber, A. B. (2011).Npn-1 contributes to axon-axon interactions that differentially control sensory andmotor innervation of the limb. PLoS Biol. 9, e1001020. doi:10.1371/journal.pbio.1001020

Ji, S.-J. and Jaffrey, S. R. (2012). Intra-axonal translation of SMAD1/5/8 mediatesretrograde regulation of trigeminal ganglia subtype specification. Neuron 74,95-107. doi:10.1016/j.neuron.2012.02.022

Kania, A., Johnson, R. L. and Jessell, T. M. (2000). Coordinate roles for LIMhomeobox genes in directing the dorsoventral trajectory of motor axons in thevertebrate limb. Cell 102, 161-173. doi:10.1016/S0092-8674(00)00022-2

Kramer, I., Sigrist, M., De Nooij, J. C., Taniuchi, I., Jessell, T. M. and Arber, S.(2006). A role for Runx transcription factor signaling in dorsal root ganglionsensory neuron diversification. Neuron 49, 379-393. doi:10.1016/j.neuron.2006.01.008

Lallemend, F. and Ernfors, P. (2012). Molecular interactions underlying thespecification of sensory neurons. Trends Neurosci. 35, 373-381. doi:10.1016/j.tins.2012.03.006

Lallemend, F., Sterzenbach, U., Hadjab-Lallemend, S., Aquino, J. B., Castelo-Branco, G., Sinha, I., Villaescusa, J. C., Levanon, D., Wang, Y., Franck, M. C.et al. (2012). Positional differences of axon growth rates between sensoryneurons encoded by runx3. EMBO J. 31, 3718-3729. doi:10.1038/emboj.2012.228

Lentz, S. I., Knudson, C. M., Korsmeyer, S. J. and Snider, W. D. (1999).Neurotrophins support the development of diverse sensory axon morphologies.J. Neurosci. 19, 1038-1048. doi:10.1523/JNEUROSCI.19-03-01038.1999

Levanon, D., Bettoun, D., Harris-Cerruti, C., Woolf, E., Negreanu, V., Eilam, R.,Bernstein, Y., Goldenberg, D., Xiao, C., Fliegauf, M. et al. (2002). The Runx3transcription factor regulates development and survival of TrkC dorsal root ganglianeurons. EMBO J. 21, 3454-3463. doi:10.1093/emboj/cdf370

Mears, S. C. and Frank, E. (1997). Formation of specific monosynaptic connectionsbetweenmuscle spindle afferents andmotoneurons in themouse. J. Neurosci. 17,3128-3135. doi:10.1523/JNEUROSCI.17-09-03128.1997

Mendelsohn, A. I., Simon, C. M., Abbott, L. F., Mentis, G. Z. and Jessell, T. M.(2015). Activity regulates the incidence of heteronymous sensory-motorconnections. Neuron 87, 111-123. doi:10.1016/j.neuron.2015.05.045

Mendelson, B. and Frank, E. (1991). Specific monosynaptic sensory-motorconnections form in the absence of patterned neural activity and motoneuronalcell death. J. Neurosci. 11, 1390-1403. doi:10.1523/JNEUROSCI.11-05-01390.1991

Nakamura, S., Senzaki, K., Yoshikawa, M., Nishimura, M., Inoue, K., Ito, Y.,Ozaki, S. and Shiga, T. (2008). Dynamic regulation of the expression ofneurotrophin receptors by Runx3. Development 135, 1703-1711. doi:10.1242/dev.015248

Oliveira, A. L. R., Hydling, F., Olsson, E., Shi, T., Edwards, R. H., Fujiyama, F.,Kaneko, T., Hokfelt, T., Cullheim, S. andMeister, B. (2003). Cellular localizationof three vesicular glutamate transporter mRNAs and proteins in rat spinal cord anddorsal root ganglia. Synapse 50, 117-129. doi:10.1002/syn.10249

Ozaki, S. and Snider, W. D. (1997). Initial trajectories of sensory axons towardlaminar targets in the developing mouse spinal cord. J. Comp. Neurol. 380,215-229. doi:10.1002/(SICI)1096-9861(19970407)380:2<215::AID-CNE5>3.0.CO;2-6

Patel, T. D., Kramer, I., Kucera, J., Niederkofler, V., Jessell, T. M., Arber, S. andSnider, W. D. (2003). Peripheral NT3 signaling is required for ETS proteinexpression and central patterning of proprioceptive sensory afferents. Neuron 38,403-416. doi:10.1016/S0896-6273(03)00261-7

Pecho-Vrieseling, E., Sigrist, M., Yoshida, Y., Jessell, T. M. andArber, S. (2009).Specificity of sensory-motor connections encoded by Sema3e-Plxnd1recognition. Nature 459, 842-846. doi:10.1038/nature08000

Poliak, S., Norovich, A. L., Yamagata, M., Sanes, J. R. and Jessell, T. M. (2016).Muscle-type identity of proprioceptors specified by spatially restricted signals fromlimb mesenchyme. Cell 164, 512-525. doi:10.1016/j.cell.2015.12.049

Price, S. R., De Marco Garcia, N. V., Ranscht, B. and Jessell, T. M. (2002).Regulation of motor neuron pool sorting by differential expression of type IIcadherins. Cell 109, 205-216. doi:10.1016/S0092-8674(02)00695-5

Sherrill, H. E., Jean, P., Driver, E. C., Sanders, T. R., Fitzgerald, T. S., Moser, T.and Kelley, M. W. (2019). Pou4f1 defines a subgroup of Type I spiral ganglionneurons and is necessary for normal inner hair cell presynaptic Ca(2+) signaling.J. Neurosci. 39, 5284-5298. doi:10.1523/JNEUROSCI.2728-18.2019

Surmeli, G., Akay, T., Ippolito, G. C., Tucker, P. W. and Jessell, T. M. (2011).Patterns of spinal sensory-motor connectivity prescribed by a dorsoventralpositional template. Cell 147, 653-665. doi:10.1016/j.cell.2011.10.012

Tourtellotte, W. G., Keller-Peck, C., Milbrandt, J. and Kucera, J. (2001). Thetranscription factor Egr3 modulates sensory axon-myotube interactions duringmuscle spindle morphogenesis. Dev. Biol. 232, 388-399. doi:10.1006/dbio.2001.0202

Usoskin, D., Furlan, A., Islam, S., Abdo, H., Lonnerberg, P., Lou, D., Hjerling-Leffler, J., Haeggstrom, J., Kharchenko, O., Kharchenko, P. V. et al. (2015).Unbiased classification of sensory neuron types by large-scale single-cell RNAsequencing. Nat. Neurosci. 18, 145-153. doi:10.1038/nn.3881

Vrieseling, E. and Arber, S. (2006). Target-induced transcriptional control ofdendritic patterning and connectivity in motor neurons by the ETS gene Pea3.Cell127, 1439-1452. doi:10.1016/j.cell.2006.10.042

Wang, Z., Li, L. Y., Taylor, M. D., Wright, D. E. and Frank, E. (2007). Prenatalexposure to elevated NT3 disrupts synaptic selectivity in the spinal cord.J. Neurosci. 27, 3686-3694. doi:10.1523/JNEUROSCI.0197-07.2007

Wenner, P. and Frank, E. (1995). Peripheral target specification of synapticconnectivity of muscle spindle sensory neurons with spinal motoneurons.J. Neurosci. 15, 8191-8198. doi:10.1523/JNEUROSCI.15-12-08191.1995

Wu, D., Schieren, I., Qian, Y., Zhang, C., Jessell, T. M. and de Nooij, J. C. (2019).A role for sensory end organ-derived signals in regulating muscle spindleproprioceptor phenotype. J. Neurosci. 39, 4252-4267. doi:10.1523/JNEUROSCI.2671-18.2019

Yang, X., Arber, S., William, C., Li, L., Tanabe, Y., Jessell, T. M., Birchmeier, C.and Burden, S. J. (2001). Patterning of muscle acetylcholine receptor geneexpression in the absence of motor innervation.Neuron 30, 399-410. doi:10.1016/S0896-6273(01)00287-2

Zhong, J., Li, X., McNamee, C., Chen, A. P., Baccarini, M. and Snider, W. D.(2007). Raf kinase signaling functions in sensory neuron differentiation and axongrowth in vivo. Nat. Neurosci. 10, 598-607. doi:10.1038/nn1898

Zhou, X., Wang, L., Hasegawa, H., Amin, P., Han, B.-X., Kaneko, S., He, Y. andWang, F. (2010). Deletion of PIK3C3/Vps34 in sensory neurons causes rapidneurodegeneration by disrupting the endosomal but not the autophagic pathway.Proc. Natl. Acad. Sci. USA 107, 9424-9429. doi:10.1073/pnas.0914725107

10

RESEARCH ARTICLE Development (2019) 146, dev181750. doi:10.1242/dev.181750

DEVELO

PM

ENT