Plexin signaling selectively regulates the stereotypedpruning of corticospinal axons from visual cortexLawrence K. Low*, Xiao-Bo Liu*†, Regina L. Faulkner*, Jeffrey Coble*, and Hwai-Jong Cheng*‡§

*Center for Neuroscience, †Department of Cell Biology and Human Anatomy, and ‡Department of Neurobiology, Physiology, and Behavior and Departmentof Pathology and Laboratory Medicine, University of California, Davis, CA 95618

Communicated by Edward G. Jones, University of California, Davis, CA, April 21, 2008 (received for review January 27, 2008)

Neurons in the developing CNS tend to send out long axoncollaterals to multiple target areas. For these neurons to attainspecific connections, some of their axon collaterals are subse-quently pruned—a process called stereotyped axon pruning. Oneof the most striking examples of stereotyped pruning in the CNS isthe pruning of corticospinal tract (CST) axons. The long CSTcollaterals from layer V neurons of the visual and motor cortices aredifferentially pruned during development. Here we demonstratethat select plexins and neuropilins, which serve as coreceptors forsemaphorins, are expressed in visual cortical neurons at the timewhen CST axon collaterals are stereotypically pruned. By analyzingmutant mice, we find that the pruning of visual, but not motor, CSTaxon collaterals depends on plexin-A3, plexin-A4, and neuropilin-2.Expression pattern study suggests that Sema3F is a candidate localcue for the pruning of visual CST axons. Using electron microscopicanalysis, we also show that visual CST axon collaterals formsynaptic contacts in the spinal cord before pruning and that theunpruned collaterals in adult mutant mice are unmyelinated andmaintain their synaptic contacts. Our results indicate that thestereotyped pruning of the visual and motor CST axon collateralsis differentially regulated and that this specificity arises from thedifferential expression of plexin receptors in the cortex.

axon pruning � corticospinal tract � plexin

A functional nervous system depends on the precise wiring ofneuronal connections with appropriate targets. During

early development, neurons tend to send out axons with exces-sive branches to multiple target areas. When the neuronal targetsbecome mature, the unnecessary branches are specificallypruned. Stereotyped axon pruning, or pruning of long axoncollaterals in a predictable manner, is a major phenomenon inthe developing CNS. This type of pruning has been observed inspecies ranging from Drosophila to mouse and is thought to beessential for the normal development of the CNS (1–5).

One classic example of stereotyped pruning in higher verte-brates is in the developing corticospinal tract (CST) (6–10). Indeveloping rodents, CST axons originate from layer V corticalpyramidal neurons in all regions of the neocortex (9, 11). Theseaxons are guided through the internal capsule, cerebral pedun-cle, and pyramidal tract and then turn dorsally to cross themidline at the pyramidal decussation before they reach thecontralateral spinal cord (Fig. 1A). The targeting of primary CSTaxons to the spinal cord is followed by axon collateral branchingto targets in the brainstem and spinal cord (Fig. 1B). This initialprojection pattern of CST axons is later modified via stereotypedaxon pruning as regions of the neocortex become specialized,and the rostral–caudal location of parent cells within the neo-cortex determines which axon collaterals are pruned (Fig. 1C).Thus, motor neurons in rostral cortex prune their axons from thesuperior and inferior colliculi, whereas visual neurons in caudalcortex prune their axons from the inferior colliculus and spinalcord (9, 12). Several neuronal cell-type specification genes suchas Ctip2 and Fezl have been shown to play roles in the develop-ment of layer V cortical neurons, but these genes seem to berequired irrespective of the locations of these neurons in the

neocortex (13–15). The transcription factor Otx1 has beenimplicated in stereotyped pruning of the visual CST (16).However, whether the pruning of exuberant CST axon collat-erals is directed by local signals or is preprogrammed is notknown. Because the differential pruning of visual and motorCST axon collaterals occurs simultaneously, it is also important

Author contributions: L.K.L., X.-B.L., and R.L.F. contributed equally to this work; L.K.L.,X.-B.L., R.L.F., and H.-J.C. designed research; L.K.L., X.-B.L., R.L.F., and J.C. performedresearch; L.K.L., X.-B.L., R.L.F., and J.C. analyzed data; and L.K.L., X.-B.L., R.L.F., and H.-J.C.wrote the paper.

The authors declare no conflict of interest.

§To whom correspondence should be addressed at: Center for Neuroscience, University ofCalifornia, 1544 Newton Court, Davis, CA 95618. E-mail: hjcheng@ucdavis.edu.

This article contains supporting information online at www.pnas.org/cgi/content/full/0803849105/DCSupplemental.

© 2008 by The National Academy of Sciences of the USA

Fig. 1. Plexin and neuropilin expression in the neocortex during CST prun-ing. (A–C) Diagrams of sagittal views of the brain representing different stagesof CST development: axon targeting by P0, axon branching between P3 andP7, and stereotyped axon pruning between P10 and P14. (D and E) PLXA3 andPLXA4 mRNAs are expressed throughout the cortex at P7, and the expressionbecomes restricted to visual cortex by P11 (black arrowheads). (F) NPN-1 mRNAis not expressed in the neocortex between P7 and P15. (G) NPN-2 mRNA isexpressed in the superficial and deep layers (red arrowheads) of the visualcortex between P7 and P15. D, dorsal; C, caudal; IC, inferior colliculus; MC,motor cortex; Pn, pons; Pyr Dec, pyramidal decussation; SC, superior colliculus;SpC, spinal cord; VC, visual cortex. (Scale bars: 1,000 �m.)

8136–8141 � PNAS � June 10, 2008 � vol. 105 � no. 23 www.pnas.org�cgi�doi�10.1073�pnas.0803849105

to know whether the two pruning processes are controlled by thesame mechanisms.

Here we report that semaphorin signaling through plexin-A3(PLXA3) and plexin-A4 (PLXA4) regulates the stereotypedpruning of the visual CST. The plexins are a family of axonguidance molecules that serve as receptors for semaphorinligands (17–19). Most secreted semaphorins (class 3) interactwith plexins through the coreceptors neuropilin-1 (NPN-1) andneuropilin-2 (NPN-2), whereas membrane-bound semaphorins(classes 4–7) can interact directly with plexins (18). Semaphorinsignaling through plexins has been associated with severalaspects of neuronal development, including stereotyped pruningof the infrapyramidal bundle in the hippocampus (17–21). Wefind that PLXA3, PLXA4, and NPN-2 are required for thestereotyped removal of visual, but not motor, CST axon collat-erals during postnatal development. We also find that Sema3F isexpressed specifically in the dorsal spinal cord and inferiorcolliculus and may interact with the plexin and neuropilincoreceptors to initiate visual CST axon pruning.

ResultsThe Expression of Plexins and Neuropilins in Layer V Cortical NeuronsCoincides with Visual CST Axon Pruning. Previous studies haveshown that Sema3F signaling through PLXA3, PLXA4, andNPN-2 regulates the stereotyped pruning of the infrapyramidalbundle in the hippocampus (20, 22–25). To address whethersemaphorin signaling through PLXA3 and PLXA4 could alsoregulate the pruning of the CST, we analyzed the mRNAexpression patterns of PLXA3 and PLXA4, as well as theneuropilins NPN-1 and NPN-2, in the neocortex. Because exu-berant CST axon collaterals are pruned in the second week ofpostnatal development (4, 11, 12, 26), we focused our expressionstudies on this time window.

PLXA3 and PLXA4 were expressed throughout the cortex atpostnatal day 7 (P7) (Fig. 1 D1 and E1). Between P7 and P11,over approximately the same time window at which stereotypedpruning of visual CST axons begins, their expression becamerestricted to the visual cortex (Fig. 1 D2 and E2). The levels ofPLXA3 and PLXA4 expression were down-regulated thereafter(Fig. 1 D3 and E3). In addition, the expression of NPN-2 waselevated in layers II/III and V of the visual cortex after P7 [Fig.1G and supporting information (SI) Fig. S1O], which coincideswith the restricted expression of PLXA3 and PLXA4 in the visualcortex. In contrast, no expression of NPN-1 in the neocortex wasdetected in the second postnatal week (Fig. 1F).

CST projections arise predominantly from type I layer V corticalneurons (9, 27), which specifically express Ctip2 (13). PLXA3 andPLXA4 are broadly expressed in all layers of the visual cortex, andwe found that a majority of Ctip2-immunopositive visual pyramidalneurons coexpressed mRNA for NPN-2 at P7 (Fig. S1O). Thesedata suggest that plexins and neuropilins could regulate the pruningof visual CST collaterals. To test this, we investigated whether thepruning of CST axons was affected in mutant mice. As a control,we first explored whether regional as well as layer-specific pattern-ing of the neocortex were altered in PLXA3 and PLXA4 mutantmice (throughout the text, PLXA3�/�, PLXA4�/�, and PLXA3/PLXA4�/� indicate PLXA3 and PLXA4 single and double knock-outs, respectively), and we observed no patterning defects inPLXA3/PLXA4�/� mice compared with WT animals (n � 2;Fig. S1).

PLXA3 and PLXA4 Are Required for the Stereotyped Pruning of VisualCST Axon Collaterals. Given the specific expression of PLXA3 andPLXA4 in the visual cortex after P7, we predicted that these twogenes participate in the pruning of visual CST axons from thespinal cord and inferior colliculus, but not of motor CST axonsfrom the superior and inferior colliculi. To test this hypothesis,we performed retrograde and anterograde tracing experiments

in mutant mice. We injected the retrograde tracer cholera toxinsubunit b (CTB) into the dorsal cervical spinal cord of P8 WTmice before pruning (n � 3) and P25 WT (n � 3) andPLXA3/PLXA4�/� (n � 4) mice after pruning to localize thedistribution of retrogradely labeled layer V neurons in the cortex(Fig. 2A). CTB-labeled cells were visible in layer V of the entireneocortex of P8 WT mice but were absent from the visual cortexof P25 WT mice (Fig. 2 B, C, E, and F), demonstrating that layerV visual neurons prune their projections from the spinal cord byP25. In contrast, CTB-labeled neurons in layer V were visiblethroughout the cortex of P25 PLXA3/PLXA4�/� mice (Fig. 2 Dand G), indicating that neurons in the visual cortex fail to prunetheir axons from the spinal cord in P25 PLXA3/PLXA4�/� mice.

To study this aberrant CST projection in PLXA3/PLXA4�/�

mice further, we performed anterograde tracing experiments inWT and mutant mice during and after pruning. When we injected1,1�-dioctadecyl-3,3,3�,3�-tetramethylindocarbocyanine perchlor-ate (DiI) into the visual cortex of WT mice (Fig. 2H), we found thatCST axons were in the process of pruning at P13 (n � 4; Fig. 2I)and that pruning had been completed by P25 (n � 4; Fig. 2 J). Incontrast, when we analyzed P25 PLXA3/PLXA4�/� mice with DiIinjections in visual cortex (n � 4; Fig. 2K), we observed a largebundle of axons that extended beyond the rostral pons (Fig. 2L) andterminated at cervical and upper thoracic spinal cord levels (Fig.2M). In addition, we observed numerous unpruned axons in theinferior colliculus of all P25 PLXA3/PLXA4�/� mice analyzed (n �4; data not shown). However, in P25 WT control mice, some axonallabeling was also noted in the inferior colliculus (n � 2 of four mice).The number of labeled axons in the inferior colliculus of WT micewas visibly less than that in mutant mice. Because of the anatomicalproximity of the visual and auditory cortices, the required amountof tracer injection into the visual cortex for quantification oftenresulted in a significant spillover of the tracer to the neighboringauditory cortex, which normally targets the inferior colliculus.Because of this limitation, we were unable to reach a quantitativeconclusion, but the analysis is indicative of a pruning defect fromthe inferior colliculus of PLXA3/PLXA4�/� mice as well.

We next analyzed the visual CST axons in PLXA3/PLXA4�/�

mice before pruning to assess whether the growth and pruningof the axons was simply delayed. At P7, PLXA3/PLXA4�/� micewith injections of DiI into the visual cortex have visual CSTaxons in the pyramidal decussation and spinal cord as seen inWT mice (Fig. 2 N–P). This indicates that the growth andpruning of visual CST axons are not simply delayed, but thatpruning is truly defective in PLXA3/PLXA4�/� mice.

To analyze the respective contributions of PLXA3 and PLXA4for the pruning of visual CST axons from the spinal cord in vivo,we injected biotinylated dextran amine (BDA) into the visualcortex of WT (n � 3), PLXA3�/� (n � 3), PLXA4�/� (n � 4),and PLXA3/PLXA4�/� (n � 3) mice older than P25 to antero-gradely label the CST axons. Because the visual CST axons inWT adult animals are pruned back to the rostral pons, wequantified the severity of the defect by obtaining counts for therelative number of axons that progressed beyond the rostral ponsin mutant mice (Fig. 2Q; see Methods). No significant differenceswere observed between WT and PLXA3�/� mice. However,PLXA4�/� mice exhibited significantly higher relative numbersof axons beyond the rostral pons compared with WT mice. Inaddition, the pruning defect was more severe in PLXA3/PLXA4�/� mice (Fig. 2Q), indicating that both PLXA3 andPLXA4 contribute to the pruning of visual CST axons from thespinal cord but that PLXA4 is preferentially required in vivo.

PLXA3 and PLXA4 Are Not Required for the Stereotyped Pruning ofMotor CST Axon Collaterals. To test whether PLXA3 and PLXA4regulate the stereotyped pruning of motor CST axons, weinjected green fluorescent microsphere (GFM) retrograde trac-ers into the superior colliculus to examine the distribution of

Low et al. PNAS � June 10, 2008 � vol. 105 � no. 23 � 8137

NEU

ROSC

IEN

CE

layer V motor neurons that extend transient axon collaterals tothe superior colliculus (Fig. 3A). In WT mice, layer V neuronsin rostral and caudal regions of the cortex were retrogradelylabeled with GFM at P6 (n � 3; Fig. 3 B and E). However, at P25(n � 4) only layer V cells in the caudal half of the cortex werelabeled (Fig. 3 C and F). In PLXA3/PLXA4�/� mice (n � 3),GFM-labeled layer V cells were absent from more rostral regionsof the cortex at P25, as we had observed in WT mice (Fig. 3 Dand G). In addition, we carried out anterograde injections ofBDA in the motor cortex of WT (n � 3) and PLXA3/PLXA4�/�

(n � 3) mice at P20 and observed no BDA-labeled axons in thecolliculi in these mice (data not shown). Thus, both anterogradeand retrograde labeling results confirmed that stereotyped prun-ing of motor CST axon collaterals from the midbrain was normalin PLXA3/PLXA4�/� mice.

NPN-2 Is Required for the Stereotyped Pruning of Visual CST AxonCollaterals. An increase in NPN-2 expression in the visual cortexafter P7 (Fig. 1G and Fig. S1O) implies that NPN-2 may interactwith PLXA3 and PLXA4 to mediate visual CST axon pruningthrough semaphorin signals. An analysis of BDA-injectedNPN-2 mutant (NPN-2�/�) mice at P25 and older (n � 4)revealed a large percentage of visual CST axons that extendedbeyond the rostral pons and into the spinal cord (Fig. 4A).

NPN-2�/� pruning defects were as severe as those observed inPLXA3/PLXA4�/� mice, suggesting that NPN-2 regulated ste-reotyped pruning by serving as a coreceptor for both plexinreceptors (Fig. 2Q).

We next examined the expression patterns of all class 3semaphorins that could potentially interact with NPN-2 in thedorsal spinal cord. We found that only Sema3F mRNA wasexpressed in the superficial one-third of the dorsal spinal cordfrom P3–P15 in WT mice, suggesting that Sema3F is involved ininitiating the pruning of visual CST axons (Fig. 4 B and C anddata not shown). We also discovered that Sema3F was expressedin the inferior colliculus, another transient target of visual CSTaxon collaterals (Fig. 4B and data not shown). In contrast,Sema3F is not expressed in the superior colliculus where visualCST axons are retained (Fig. 4B). Taken together, our resultssuggest that NPN-2 serves as a coreceptor to PLXA3 and PLXA4in CST axon remodeling in the spinal cord and suggest a possiblerole of Sema3F as a candidate ligand for these receptors.

Unpruned Visual CST Axon Collaterals in PLXA3/PLXA4�/� Mice AreUnmyelinated and Retain Their Synaptic Contacts in the Spinal Cord.We have previously shown that hippocampal mossy fibers formsynaptic contacts with their targets before pruning (24). How-ever, it is still unclear whether transient CST axons in the spinal

Fig. 2. PLXA3 and PLXA4 are required for the stereotyped pruning of visual corticospinal axons from the spinal cord. (A–G) Layer V neurons in the visual cortex(orange) were retrogradely labeled by injecting CTB into the dorsal cervical spinal cord (red arrowhead in A). The distribution of labeled neurons in the visualcortex is normalized for comparison as diagramed in A (a/b ratio; see Methods). A Right shows that visual CST axons are not pruned from the spinal cord of P25PLXA3/PLXA4�/� (DKO) mice (mean � SEM; n values are indicated in parentheses). **, P � 0.01 (Student’s t test). Representative images of retrogradely labeledneurons in the cortex for each set of experiments are shown in B–D. The Inset in each image summarizes the distribution of CTB-positive neurons (red) in thecortex, and the white arrowheads indicate the distribution of neurons in the visual cortex. Higher-magnification images taken from boxed regions in B–D areshown in E–G, respectively. (H–P) Anterograde tracing of axons from the visual cortex in WT and DKO mice. Red arrowheads in H, K, and N indicate the injectionsites. DiI-labeled axons in WT mice are in the process of pruning at P13 (arrows in I), and at P25 axons terminate in the rostral pons (J). DiI-labeled axons in P25DKO animals extend beyond the rostral pons (L), cross to the contralateral spinal cord at the pyramidal decussation, and terminate in the spinal cord (M).Unpruned axons in the DKO spinal cord (arrow) are shown in higher magnification in (M�). DiI-labeled axons in P7 DKO animals (arrows) grow normally into thepyramidal decussation (O) and dorsal spinal cord (P). White dashes in I, J, L, M, and P) indicate meninges and do not represent positive DiI labeling. (Q) Acomparison of normalized ratios of BDA-labeled visual CST axons (mean � SEM) present at the pons, brainstem, pyramidal decussation (Pyr Dec), and cervicalspinal cord (Cerv SpC) in mice aged P30–P35 (see Methods). When compared with WT mice, significant pruning defects are found in PLXA4�/�, DKO, and NPN-2�/�

mice [*, P � 0.05 (ANOVA, Newman–Keuls test)], but not in PLXA3�/� mice. The defects in DKO and NPN-2�/� mice are also more severe than in PLXA3�/� orPLXA4�/� mice [*, P � 0.05 (ANOVA, Newman–Keuls test)] at all levels beyond rostral pons. n values are indicated in parentheses. dSpC, dorsal spinal cord; Pn,pons. [Scale bars: 1,000 �m (B–D) and 200 �m (E–G, I, J, L, M, O, and P).]

8138 � www.pnas.org�cgi�doi�10.1073�pnas.0803849105 Low et al.

cord of WT mice form synaptic contacts with their targets. UsingBDA axon tracing combined with EM, we analyzed the devel-opment of transient CST axons from the visual cortex of WTmice (Fig. 5A). BDA-labeled CST axons from the visual cortexwere easily visible in the dorsal spinal cord by P7 (Fig. 5B). Thesetransient CST axons were unmyelinated (data not shown), as arethe majority of CST axons from motor cortex at this time (6, 9),and exhibited branches that terminated in bouton-like structuresin the gray matter of the spinal cord (Fig. 5B). Serial sectionanalysis of these boutons revealed that several (n � 5 of nineboutons) established asymmetric synaptic contacts with theirtargets (Fig. 5 C and D). Three-dimensional EM reconstructionof these boutons (n � 3) demonstrated that the perimeter oftransient boutons and the length of synapses were consistentwith those of motor CST axons in the gray matter at comparableages (Fig. 5E). The formation of synaptic contacts suggeststhat transient axons from the visual cortex have the potentialto communicate with these targets before stereotyped axonpruning.

To examine the development of unpruned visual CST axons inmutant mice, we analyzed BDA-labeled visual CST axons in thecaudal pons and pyramidal tract of P30 PLXA3/PLXA4�/� mice(n � 3) using EM. All BDA-labeled unpruned visual CST fiberswere found to be unmyelinated (n � 8 BDA-labeled axons) (Fig.

5J) and were surrounded by large numbers of BDA-immunonegative, myelinated axons. A majority of motor CSTaxons were found to be myelinated in WT (�90%) and PLXA3/PLXA4�/� (�90%) mice, suggesting that a large percentage ofthe unpruned visual CST axons failed to develop normally. Thediameter of the BDA-labeled, unmyelinated axons was signifi-cantly smaller than that of surrounding BDA-negative, myelin-ated axons in both the pons and pyramidal tract (Fig. 5M).Despite the unusual absence of myelin in unpruned axons,several were found to extend branches within the gray matter ofthe spinal cord and terminate in bouton-like expansions (Fig. 5F and G). Three-dimensional EM reconstruction of these bou-tons and quantification of electron micrographs (n � 8 boutons)demonstrated that they were comparable in size to motor CSTterminations in the spinal cord (Fig. 5K). Furthermore, we foundthat nearly half (n � 3 of seven boutons) of the unpruned visualCST boutons still made asymmetric synaptic contacts on den-dritic profiles (Fig. 5 H and I). The number of synaptic contactsper bouton was significantly lower for unpruned visual CSTaxons in PLXA3/PLXA4�/� mice compared with motor CSTboutons in WT animals (Fig. 5L). In summary, our resultsdemonstrate that although nearly 50% of unpruned visualCST axons retain synapses in the spinal cord, these unprunedaxons are likely to be physiologically impaired because theyare unmyelinated.

DiscussionThe development of the CST has served as a classical examplefor studying the stereotyped pruning of long-range axon collat-erals (3, 4, 9). Our analysis of this system in mutant mice hasidentified specific roles for plexin/neuropilin signaling in con-trolling the stereotyped pruning of visual CST axons. We foundthat the pruning of visual CST collaterals from the spinal corddepends directly on PLXA3, PLXA4, and NPN-2 signaling.Sema3F is the only class 3 semaphorin expressed in the targetsof visual CST axons before and during stereotyped axon pruning.Thus, an increased sensitivity to Sema3F within local targets ofvisual CST axon collaterals could initiate pruning (Fig. S2). Thisfinding corroborates previous reports showing that this same setof ligand and receptors is required for the stereotyped pruningof hippocampal mossy fiber collaterals (20, 22–25). Quantitative

Fig. 3. PLXA3 and PLXA4 are not required for the stereotyped pruning ofmotor corticospinal axons from the superior colliculus. (A) Layer V neurons inthe motor cortex (blue) were retrogradely labeled by injection of GFM in thesuperior colliculus (green arrowhead). The distribution of labeled neurons inthe cortex is normalized for comparison as diagramed (a/b ratio; see Methods).The bar graph indicates no significant motor CST pruning defects from thesuperior colliculus of PLXA3/PLXA4�/� (DKO) mice (mean � SEM; n values areindicated in parentheses). Representative images of retrogradely labeledneurons in the cortex for each set of experiments are shown in B–D. The Insetin each image summarizes the distribution of CTB-positive neurons (green) inthe cortex. Layer V motor neurons that extend transient projections to thesuperior colliculus at P6 are indicated by white arrowheads (B). Yellow arrow-heads (C and D) mark the border of GFM-positive neurons. Higher-magnification images taken from boxed regions in B–D are shown in E–G,respectively. [Scale bars: 1,000 �m (B–D) and 200 �m (E–G).]

Fig. 4. Stereotyped pruning of the visual CST is likely to be initiated bySema3F signaling through PLXA3 and PLXA4 and their coreceptor, NPN-2. (A)Diagram and image showing unpruned BDA-positive visual CST axons (blackarrows) in the pyramidal decussation (Pyr Dec) and dorsal cervical spinal cord(dSpC) of P25 NPN-2�/� mice. (B) Sema3F mRNA expression is observed in thetransient targets of visual CST axon collaterals at the inferior colliculus (greenarrowheads) and dorsal spinal cord (red arrowheads) of P7 and P11 WT mice.A line is shown separating the superior and inferior colliculus. Sema3F mRNAis absent from the superior colliculus (blue arrowhead), which retains its visualCST axon collaterals. Dashed lines indicate the regions from which transversesections in C are taken. (C) Sema3F mRNA expression in the dorsal regions oftransverse sections of the spinal cord at P7 and P11 (red arrowheads). Dashedlines outline the dorsal funiculus, where CST axons are located. DF, dorsalfuniculus; IC, inferior colliculus; SC, superior colliculus; SpC, spinal cord. [Scalebars: 100 �m (A), 1,000 �m (B), and 500 �m (C).]

Low et al. PNAS � June 10, 2008 � vol. 105 � no. 23 � 8139

NEU

ROSC

IEN

CE

analysis of the CST pruning defect in mutants revealed thatPLXA3 and PLXA4 are differentially required in vivo and thatPLXA4 plays a larger role in this system than PLXA3. We haveidentified similar preferential uses of these two plexins inregulating the in vivo guidance of peripheral sensory and sym-

pathetic axons (28, 29), implying that differential usages ofcoexisting plexins are common in vivo. With the specific anddiverse expression patterns of members of the semaphorin ligandand receptor families in the developing nervous system, ourresults further support the notion that semaphorin signalingplays essential roles in mediating stereotyped pruning through-out the CNS.

We also demonstrate that the local signals for the pruning ofaxon collaterals from visual and motor CST are distinct, eventhough the early projection patterns from the two corticalregions are indistinguishable. This specificity depends on thespatial and temporal expression patterns of plexins and neuro-pilins in the visual cortex, as well as semaphorin in the relevantsubcortical regions during stereotyped axon pruning. The ho-meodomain transcription factor Otx1 was previously found to beexpressed within all layer V cortical projection neurons duringthe pruning of CST axon collaterals (16). Because loss of Otx1expression in mutant mice results in visual CST pruning defectswithin the spinal cord and midbrain, it will be interesting todetermine whether Otx1 directly regulates the expression ofPLXA3, PLXA4, and NPN-2 in the visual cortex. In addition,what factors could account for the pruning of motor CST axoncollaterals from the superior and inferior colliculi? One possi-bility is that transcription factors related to Otx1 may determinethe expression of other plexins or axon guidance receptors thattrigger the specific pruning of motor CST axon collaterals.Alternatively, these motor collaterals may lack the receptors thatallow axons to sense stabilizing trophic factors in the colliculi.

Visual CST axons form transient synaptic contacts with theirtargets in the spinal cord before pruning. When axonal remod-eling is impaired in mutant mice, the synapses are retained.Unpruned visual CST axons in the spinal cords of plexin mutantsare smaller in size and make fewer synaptic contacts than motorCST axons. In addition, these unpruned axons are unmyelinatedeven though CST axons from motor cortex are generally my-elinated in adult mice (6, 9). This unexpected finding suggeststhat other factors in addition to synaptic contacts, and thusneuronal activity, may be required for the proper myelination ofthe CST axons. It is plausible that several of these abnormalaxonal branches and synaptic contacts in plexin mutants may bephysiologically impaired, which might result in pathologicalchanges later in adulthood. Certainly, more studies will beneeded to address how the absence of plexin signaling affects thephysiology of visual and motor responses and what pathologicalchanges will occur because of the existence of the abnormalconnections. Abnormal pruning of axonal branches has beenimplied in neurodevelopmental disorders; it will be of interest toexplore whether changes that occur to the unpruned axoncollaterals throughout the life of the plexin/neuropilin mutantsare similar to the pathological findings observed in patients withneuropsychiatric developmental disorders such as schizophreniaand autism.

MethodsMouse Breeding. Animal protocols were approved by the Institutional AnimalCare and Use Committee at the University of California, Davis. Genotyping onknockout mice was carried out as described previously (23, 29, 30).

Mouse Tracer Injections. WT and mutant mice were injected with varioustracers at different postnatal ages (P0–P45). DiI (Molecular Probes) and BDA(Molecular Probes) anterograde tracing was performed as described previ-ously (31, 32). Mice were injected blindly before determining genotype.Briefly, focal injections of DiI (20% in N,N-dimethylformamide) or BDA (10–20% in PBS) were made in either the motor or occipital cortex of WT andmutant mice in vivo and allowed to trace for a minimum of 3 days. Locationsof the injection sites were confirmed in sagittal sections of the cortex to ensurethat tracers were injected in the appropriate regions of the cortex.

For retrograde tracing studies, GFM (Lumafluor) and Alexa Fluor 594-conjugated CTB (Molecular Probes) were injected into the superior colliculus

Fig. 5. EM analysis of visual CST axons in the spinal cord of WT andPLXA3/PLXA4�/� mice. (A) Diagram of a transverse cervical spinal cord sectionshowing visual CST axons branching ventrally into the gray matter. The boxindicates the location where the images in B, F, and G are taken. (B) ABDA-labeled transient visual CST axon in P7 WT mouse branches into thecervical spinal cord gray matter and exhibits bouton-like structures (redarrowheads). (C and D) Serial electron micrographs showing a BDA-labeledvisual CST axon terminal (t) forming an asymmetric synapse (black arrowhead)with a dendrite (d) in the gray matter of the cervical spinal cord of P7 WTmouse. (E) Three-dimensional EM reconstruction of serial sections from aBDA-labeled terminal shown in C and D demonstrating an axon terminal(green) adjacent to a postsynaptic density (yellow). (F and G) UnprunedBDA-labeled visual CST axons are found in the dorsal CST of the spinal cord ofP30 PLXA3/PLXA4�/� (DKO) mice. Axon branches within the gray mattercontain bouton-like structures (red arrowheads). (H) An electron micrographshowing a BDA-positive visual CST axon terminal (t) in the gray matter of thespinal cord of a P30 DKO mouse. The BDA-labeled terminal contains vesiclesthat cluster adjacent to a postsynaptic density (black arrowhead) within adendritic spine (sp). (I) Three-dimensional reconstruction of the axon terminalin H (green) adjacent to a postsynaptic density (yellow). (J) An electronmicrograph showing unmyelinated, unpruned BDA-positive visual CST axons(a) adjacent to unlabeled myelinated axons (asterisks) within the brainstem.(K) A comparison of average bouton sizes (mean � SEM) of adult WT motorand adult DKO visual axons in the spinal cord indicates no significant differ-ences. (L) A comparison of average synapse number per bouton (mean � SEM)for adult WT motor and adult DKO visual axonal boutons in the spinal cordindicates significantly fewer synapses made by unpruned adult DKO visualaxonal boutons [*, P � 0.05 (Student’s t test)]. (M) A comparison of averageaxon diameter (mean � SEM) of BDA-positive unpruned visual axons in DKOanimals and surrounding motor axons indicates that in the caudal pons andpyramidal tract (Pyr Tract) DKO unpruned visual axons are significantly smallerin size [*, P � 0.05; **, P � 0.01 (Student’s t test)]. n values are indicated inparentheses. [Scale bars: 50 �m (B), 0.25 �m (C–E), 25 �m (F and G), 0.5 �m (H,I, and Inset in J), and 1 �m (J).]

8140 � www.pnas.org�cgi�doi�10.1073�pnas.0803849105 Low et al.

or cervical spinal cord of WT and mutant mice and allowed to trace in vivo fora minimum of 2 days. Mice injected with anterograde or retrograde tracerswere perfused with 4% paraformaldehyde, and their brains were cut sagit-tally or coronally at 50–100 �m as described previously (24).

Immunohistochemistry and in Situ Hybridization (ISH). Immunohistochemistrywas performed on free-floating sections as described previously (24). Ctip2was used as an antibody in the study (1:1,000; Abcam). The probes for ISH andthe procedures for radioactive �-33P ISH were as described previously (33).Combined immunohistochemical and ISH colocalization experiments wereperformed on tissue sections mounted on slides. Briefly, nonradioactive ISHwas performed on 12- to 14-�m cryosectioned brains by using digoxigenin-labeled probes for NPN-2 as described previously (23). Sections were thendeveloped for Ctip2 immunohistochemical colocalization by using the meth-ods described (24).

Analysis of Retrogradely Labeled Layer V Cortical Neurons. Sagittal sections ofthe brains of WT and mutant mice were used for analysis. In mice injected withCTB in the cervical spinal cord or GFM in the superior colliculus, raw images ofthe cortex were taken with a CCD camera (Zeiss) and imported into PhotoShop(Adobe Systems). Montaged images of the sagittally cut brain were created inPhotoShop and analyzed with NIH Image J. For CTB-injected mice, the rostralblade of the dentate gyrus of the hippocampus was used as an arbitrarylandmark for analysis (because most layer V neurons caudal to this regionextend transient projections to the spinal cord). The length of cortex occupiedby CTB-positive neurons caudal to the rostral blade of the dentate gyrus wasnormalized to the entire length of the occipital cortex (also measured from therostral blade of the dentate gyrus) (Fig. 2A). For GFM-injected mice, therostral–caudal length of cortex occupied by GFM-positive neurons was nor-malized to the entire length of the cortex (Fig. 3A).

Analysis of Pruning Defects for BDA-Labeled Visual Corticospinal Axons. BDAwas injected into the visual cortex of WT and mutant mice. BDA-labeled axons

were scored at different locations along the visual CST: within the rostral halfof the pons, at the brainstem rostral to the inferior olive, at the pyramidaldecussation, and within the cervical spinal cord. Axon counts were performedon all 50-�m serial sagittal sections by using an eyepiece reticule as describedin ref. 34. Labeled axons were scored only if their axon segments were �15 �mto minimize counting axons more than once in adjacent sections. Because thevisual CST is pruned back from the spinal cord to the rostral pons in WT miceby P20, the severity of the pruning defect was determined by normalizing theabsolute counts of axons at different locations along the visual CST to the totalnumber of axons accessing the rostral pons.

EM Processing and Quantification. Sections that contained BDA-labeled CSTaxons were preserved for ultrastructural analysis with EM as described (24).The perimeters of all boutons were measured in each serial section, and thenthe largest measured perimeter for each bouton was used in calculating theaverage bouton perimeter (Fig. 5K). Synapses per bouton were counted inserial sections such that each synapse was counted only once even if itappeared in multiple serial sections and was reported as synapse density (Fig.5L). For axon width measurements, axon diameters for single axons weremeasured in serial sections and then the average axon diameter for that axonwas recorded. Axon diameters were measured from electron micrographscontaining BDA-labeled axons from visual cortex of PLXA3/PLXA4�/� animals.Surrounding BDA-negative axon diameters from the same electron micro-graphs were also measured as controls for comparison (Fig. 5M). All measure-ments were done using NIH Image J.

Statistics for all data were obtained from Statistica 6.0 (Statsoft).

ACKNOWLEDGMENTS. We thank P. Nguyen, M. Chen, S. Mikula, A. Graziano,and K. Murray for technical assistance and E. G. Jones and members of theH.-J.C. and E. G. Jones laboratories for valuable discussions and comments. Thisresearch was supported by grants from the Klingenstein Fund, Autism Speaks/National Alliance for Autism Research, the March of Dimes Birth DefectsFoundation, and National Institutes of Health Grant HD045757 (to H.-J.C.).

1. Kantor DB, Kolodkin AL (2003) Curbing the excesses of youth: Molecular insights intoaxonal pruning. Neuron 38:849–852.

2. Low LK, Cheng HJ (2005) A little nip and tuck: Axon refinement during developmentand axonal injury. Curr Opin Neurobiol 15:549–556.

3. Low LK, Cheng HJ (2006) Axon pruning: An essential step underlying the developmen-tal plasticity of neuronal connections. Philos Trans R Soc London B 361:1531–1544.

4. Luo L, O’Leary DD (2005) Axon retraction and degeneration in development anddisease. Annu Rev Neurosci 28:127–156.

5. Innocenti GM, Price DJ (2005) Exuberance in the development of cortical networks. NatRev Neurosci 6:955–965.

6. Jones EG, Schreyer DJ, Wise SP (1982) Growth and maturation of the rat corticospinaltract. Prog Brain Res 57:361–379.

7. Martin JH (2005) The corticospinal system: From development to motor control.Neuroscientist 11:161–173.

8. O’Leary DD, Koester SE (1993) Development of projection neuron types, axon path-ways, and patterned connections of the mammalian cortex. Neuron 10:991–1006.

9. Stanfield BB (1992) The development of the corticospinal projection. Prog Neurobiol38:169–202.

10. Terashima T (1995) Anatomy, development and lesion-induced plasticity of rodentcorticospinal tract. Neurosci Res 22:139–161.

11. Stanfield BB, O’Leary DD (1985) The transient corticospinal projection from the occip-ital cortex during the postnatal development of the rat. J Comp Neurol 238:236–248.

12. Thong IG, Dreher B (1987) The development of the corticotectal pathway in the albinorat: Transient projections from the visual and motor cortices. Neurosci Lett 80:275–282.

13. Arlotta P, et al. (2005) Neuronal subtype-specific genes that control corticospinalmotor neuron development in vivo. Neuron 45:207–221.

14. Chen B, Schaevitz LR, McConnell SK (2005) Fezl regulates the differentiation and axontargeting of layer 5 subcortical projection neurons in cerebral cortex. Proc Natl AcadSci USA 102:17184–17189.

15. Molyneaux BJ, Arlotta P, Hirata T, Hibi M, Macklis JD (2005) Fezl is required for the birthand specification of corticospinal motor neurons. Neuron 47:817–831.

16. Weimann JM, et al. (1999) Cortical neurons require Otx1 for the refinement ofexuberant axonal projections to subcortical targets. Neuron 24:819–831.

17. He Z, Wang KC, Koprivica V, Ming G, Song HJ (2002) Knowing how to navigate:Mechanisms of semaphorin signaling in the nervous system. Sci STKE 2002:RE1.

18. Tran TS, Kolodkin AL, Bharadwaj R (2007) Semaphorin regulation of cellular morphol-ogy. Annu Rev Cell Dev Biol 23:263–292.

19. Waimey KE, Cheng HJ (2006) Axon pruning and synaptic development: How are theyper-plexin? Neuroscientist 12:398–409.

20. Faulkner RL, Low LK, Cheng HJ (2007) Axon pruning in the developing vertebratehippocampus. Dev Neurosci 29:6–13.

21. Pasterkamp RJ, Kolodkin AL (2003) Semaphorin junction: Making tracks toward neuralconnectivity. Curr Opin Neurobiol 13:79–89.

22. Bagri A, Cheng HJ, Yaron A, Pleasure SJ, Tessier-Lavigne M (2003) Stereotyped pruningof long hippocampal axon branches triggered by retraction inducers of the sema-phorin family. Cell 113:285–299.

23. Cheng HJ, et al. (2001) Plexin-A3 mediates semaphorin signaling and regulates thedevelopment of hippocampal axonal projections. Neuron 32:249–263.

24. Liu XB, Low LK, Jones EG, Cheng HJ (2005) Stereotyped axon pruning via plexinsignaling is associated with synaptic complex elimination in the hippocampus. J Neu-rosci 25:9124–9134.

25. Sahay A, Molliver ME, Ginty DD, Kolodkin AL (2003) Semaphorin 3F is critical fordevelopment of limbic system circuitry and is required in neurons for selective CNSaxon guidance events. J Neurosci 23:6671–6680.

26. Stanfield BB, O’Leary DD, Fricks C (1982) Selective collateral elimination in earlypostnatal development restricts cortical distribution of rat pyramidal tract neurones.Nature 298:371–373.

27. Molnar Z, Cheung AF (2006) Towards the classification of subpopulations of layer Vpyramidal projection neurons. Neurosci Res 55:105–115.

28. Waimey KE, Huang PH, Chen M, Cheng HJ (2008) Plexin-A3 and plexin-A4 restrict themigration of sympathetic neurons but not their neural crest precursors. Dev Biol315:448–458.

29. Yaron A, Huang PH, Cheng HJ, Tessier-Lavigne M (2005) Differential requirement forplexin-A3 and -A4 in mediating responses of sensory and sympathetic neurons todistinct class 3 semaphorins. Neuron 45:513–523.

30. Chen H, et al. (2000) Neuropilin-2 regulates the development of selective cranial andsensory nerves and hippocampal mossy fiber projections. Neuron 25:43–56.

31. Brosamle C, Schwab ME (1997) Cells of origin, course, and termination patterns of theventral, uncrossed component of the mature rat corticospinal tract. J Comp Neurol386:293–303.

32. O’Leary DD, Terashima T (1988) Cortical axons branch to multiple subcortical targets byinterstitial axon budding: Implications for target recognition and ‘‘waiting periods.’’Neuron 1:901–910.

33. Liu XB, Murray KD, Jones EG (2004) Switching of NMDA receptor 2A and 2B subunitsat thalamic and cortical synapses during early postnatal development. J Neurosci24:8885–8895.

34. Kingsbury MA, Graf ER, Finlay BL (2000) Altered development of visual subcortical pro-jections following neonatal thalamic ablation in the hamster. J Comp Neurol 424:165–178.

Low et al. PNAS � June 10, 2008 � vol. 105 � no. 23 � 8141

NEU

ROSC

IEN

CE