Long-range retrograde spread of LTP and LTD fromoptic tectum to retinaJiu-lin Dua,b, Hong-ping Weib, Zuo-ren Wanga,b, Scott T. Wonga, and Mu-ming Pooa,1

aDivision of Neurobiology, Department of Molecular and Cell Biology, Helen Wills Neuroscience Institute, University of California, Berkeley, CA 94720;and bInstitute of Neuroscience and Key State Laboratory of Neuroscience, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences,Shanghai 200031, China

This contribution is part of the special series of Inaugural Articles by members of the National Academy of Sciences elected in 2009.

Contributed by Mu-ming Poo, October 5, 2009 (sent for review July 17, 2009)

Neural activity can induce persistent strengthening or weakeningof synapses, known as long-term potentiation (LTP) or long-termdepression (LTD), respectively. As potential cellular mechanismsunderlying learning and memory, LTP and LTD are generallyregarded as synapse-specific ‘‘imprints’’ of activity, although thereis evidence in vitro that LTP/LTD may spread to adjacent synapses.Here, we report that LTP and LTD induced in vivo at retinotectalsynapses of Xenopus tadpoles undergo rapid long-range retro-grade spread from the optic tectum to the retina, resulting inpotentiation and depression of bipolar cell synapses on the den-drites of retinal ganglion cells, respectively. The retrograde spreadof LTP and LTD required retrograde signaling initiated by brain-derived neurotrophic factor and nitric oxide in the tectum, respec-tively. Such bidirectional adjustment of the strength of inputsynapses in accordance to that of output synapses may serve tocoordinate developmental refinement and learning functions ofneural circuits.

neural development � retrograde signaling � synaptic plasticity �retinal ganglion cell

Persistent modification of synaptic efficacy by neural activityrepresents a cellular mechanism underlying learning and

memory functions of the brain (1, 2). Activity-induced long-termpotentiation (LTP) and long-term depression (LTD) of synaptictransmission are found in many regions of developing andmature nervous systems (3–5). Because LTP/LTD inductionoften requires correlated presynaptic and postsynaptic activa-tion, it is generally assumed that these bidirectional modifica-tions are synapse-specific, i.e., only those synapses experiencingcorrelated presynaptic and postsynaptic activities are modified(2). This notion of synapse specificity has been widely used incomputational analysis of activity-dependent modification ofneural circuits. Experimental studies have indeed demonstratedinput specificity in the induction of LTP/LTD, when activatedand unactivated synaptic inputs are well separated (6–8).

However, many studies in the past two decades have suggestedthat this notion of input specificity needs to be modified.Induction of LTP in one pathway was found to be accompaniedby heterosynaptic depression of other pathways (9). Transmittersreleased at activated synapses may also ‘‘spill over’’ and modifyadjacent synaptic sites (10). By examining specifically the issueof input specificity, several in vitro studies have shown thatLTP/LTD induced at one set of synapses on a neuron couldspread laterally to nearby synapses (11–16). Most surprisingly, inrandom networks formed by dissociated hippocampal neurons inculture, there is a rapid retrograde spread of potentiation/depression from the site of LTP/LTD induction to the synapticinputs on the dendrite of the presynaptic neuron (17, 18). Suchlong-range presynaptic spread of retrograde signals accompa-nying LTP/LTD induction is also implicated by the findings ofrapid global modifications of intrinsic excitability of the presyn-aptic neuron after LTP/LTD induction (19, 20). However, theconfiguration of synapses formed in cultures between dissociated

neurons are quite different from that in natural neural circuits;it is thus important to determine whether long-range retrogradespread of LTP/LTD can indeed occur in intact brain tissues. Inthe present study, we have chosen the retinotectal system ofXenopus tadpoles to investigate whether LTP/LTD generated atoutput synapses of retinal ganglion cells (RGCs) in the optictectum can spread to the synapses on the dendrites of RGCs inthe retina.

Retrograde signals received by the axonal terminals areresponsible for the trophic influence of target cells on neuronalsurvival (21, 22), dendritic morphology (23, 24), and the integrityof synaptic connections on the dendrite (25, 26). Recently, wefound a rapid form of retrograde modulation in the developingXenopus visual system, application of exogenous brain-derivedneurotrophic factor (BDNF) to the RGC axon terminals in theoptic tectum triggered a retrograde signal in the RGC, resultingin the potentiation of bipolar cell (BC) synapses on RGCdendrites within 20 min (27). This study suggests that axon-to-dendrite retrograde signaling can occur rapidly and causepersistent synaptic modulation at the dendrite. Because LTPinduction is known to depend on BDNF signaling at retinotectaland hippocampal CA3–CA1 synapses (28, 29), it is possible thatLTP induced at retinotectal synapses may spread to the retina viathe action of endogenous BDNF secreted in the tectum.

In the present study, we monitored the efficacy of BC-RGCsynapses by perforated whole-cell recording from the RGC andfound that light-induced excitatory synaptic currents were in-creased and decreased within minutes after the induction of LTPand LTD at retinotectal synapses, respectively. Measurements ofthe properties of BC-RGC synapses indicated that this retro-grade potentiation/depression was caused by modification of thesingle-channel conductance and the density of postsynapticAMPA subtype of glutamate receptors (AMPARs). Further-more, we showed that BDNF and NO secreted at retinotectalsynapses as a result of LTP and LTD induction are responsiblefor triggering the retrograde potentiation and depression, re-spectively. Together, these results provide an in vivo demon-stration of long-range retrograde spread of LTP and LTD. Suchretrograde spread of LTP/LTD points to the existence ofcoordinated changes in the strength of synaptic inputs to aneuron in accordance to that of its synaptic outputs, reminiscentof the back-propagation algorithm used for supervised learningin artificial neural networks (30).

ResultsInduction of LTP and LTD at Retinotectal Synapses. In the Xenopusretinotectal system, RGCs send out axons to establish retino-

Author contributions: J.-l.D. and M.-m.P. designed research; J.-l.D., H.-p.W., and Z.-r.W.performed research; J.-l.D., Z.-r.W., and S.T.W. contributed new reagents/analytic tools;J.-l.D., Z.-r.W., S.T.W., and M.-m.P. analyzed data; and J.-l.D. and M.-m.P. wrote the paper.

The authors declare no conflict of interest.

1To whom correspondence should be addressed. E-mail: mpoo@berkeley.edu.

This article contains supporting information online at www.pnas.org/cgi/content/full/0910659106/DCSupplemental.

18890–18896 � PNAS � November 10, 2009 � vol. 106 � no. 45 www.pnas.org�cgi�doi�10.1073�pnas.0910659106

Dow

nloa

ded

by g

uest

on

Dec

embe

r 12

, 202

0

tectal synapses with tectal neurons in the contralateral optictectum and receive synaptic inputs from bipolar and amacrinecells at their dendrites in the retina. In this study, we firstexamined activity-induced LTP and LTD at retinotectal syn-apses (stages 42–44) by stimulating RGC axons at the optic nervehead and recording evoked excitatory postsynaptic currents(EPSCs) from the tectal neuron in the contralateral tectum,using perforated-patch whole-cell recording. For LTP induction,we applied five episodes (at 5-s intervals) of theta burst stimu-lation (TBS), with the tectal neuron held in the current clamp(c.c.) mode. We found that TBS induced an immediate increasein the amplitude of EPSCs that persisted for as long as therecording was made (up to 1 h; Fig. 1A). The mean EPSCamplitude during 5–45 min after TBS was 155 � 12% (SEM, n �6) of the control value obtained before TBS. There was nochange in the EPSC amplitude (104 � 7% of control, n � 5)when TBS was applied in the presence of 50 �M AP5 [D(-)-2-amino-5-phosphonovaleric acid], an antagonist of N-methyl-D-aspartate subtype of glutamate receptors (NMDARs), in therecording chamber (Fig. 1 A).

In contrast to the effect of TBS, activation of RGC axons atthe optic nerve head with low-frequency stimulation (LFS) (1 Hzfor 15 min), which is known to induce LTD at many excitatorysynapses (5), resulted in an immediate reduction in the EPSCamplitude (57 � 7% of control at 5–30 min after LFS; n � 9; Fig.1B) that persisted for as long the recording was made. This LTDalso required NMDAR activation, because it was absent (95 �4%, n � 4) when AP5 was present in the recording medium (Fig.1B). Application of LFS (1 Hz) for a shorter duration alsoinduced LTD, but the extent of depression (5-min LFS: 70 �10%, n � 3; 10-min LFS: 62 � 6%, n � 5) was smaller than thatinduced by LFS for 15 min.

Potentiation of BC-RGC Synapses After Retinotectal LTP. To examinewhether LTP/LTD induction at retinotectal synapses in the optictectum affects the efficacy of excitatory synapses made by BCs

on the dendrite of RGCs, we monitored light-evoked excitatorycompound synaptic currents (eCSCs) in the RGC by performingperforated whole-cell recording at the reversal potential ofinhibitory synaptic currents (27). Changes of eCSCs in therecorded RGC after LTP/LTD induction at retinotectal syn-apses by stimulating the optic nerve head would suggest retro-grade spread of synaptic modification. Because simultaneous invivo whole-cell recording from a tectal neuron and a RGC istechnically not yet feasible, we examine changes in eCSCs onlyin RGCs that exhibited antidromic action potentials in responseto stimulation of the optic nerve head during LTP/LTD induc-tion in all experiments to ensure that retinotectal synapses madeby the recorded RGC undergo LTP/LTD.

Whole-field illumination of the retina (2-s duration) elicitedboth ‘‘on’’ and ‘‘off’’ responses in most RGCs. The efficacy of theBC-RGC synapse was measured by the total integrated chargeassociated with the eCSC (Vc � �65 mV) within a 50-ms windowafter its onset (27). When examined at a low frequency (0.033Hz), eCSCs remained stable for up to 2 h. However, we foundthat the induction of LTP at retinotectal synapses (by TBS atoptic nerve head; Fig. 1 A) resulted in a marked enhancement ofeCSCs for both on and off responses within 5 min after TBS. Thisenhancement of eCSCs lasted for �20 min (Fig. 2A). In Fig. 2 A

Fig. 1. Induction of LTP and LTD at retinotectal synapses. (A) LTP induced byTBS. (Upper) Data from an example experiment. Changes of the EPSC ampli-tude in a tectal neuron (Vc � � 65 mV) before and after five episodes (with 5-sintervals) of TBS, each consisting of 10 trains (spaced by 200 ms) of five pulsesat 100 Hz. Dotted line is the mean value before TBS. Traces above show EPSCsat the time marked by the arrow. (Scales: 50 pA, 20 ms.) (Lower) Summary ofdata from all experiments. Filled circles indicate experiments similar to thatshown above. Open circles indicate experiments with AP5 (50 �M) in the bath.Data points were normalized by the mean value before TBS and laterallydisplaced for clarity. (B) LTD induced by LFS. Data from an example experiment(Upper) and summary of all experiments (Lower) show changes in the EPSCamplitude before and after 15 min LFS (1 Hz). (Scales: 25 pA, 20 ms.)

Fig. 2. Retrograde modulation of light-evoked excitatory responses in RGCs.(A) Enhancement of light-evoked eCSCs in RGCs induced by TBS. (Upper) Datafrom an example experiment. Data points depict the changes in the totalintegrated charge associated with eCSCs (at Vc � �65 mV) of the off responseswithin a 50-ms window after the onset of eCSCs before and after TBS (asdescribed in Fig. 1A). The RGC was held at c.c. mode during TBS. Samples ofsingle on and off eCSCs (arrow) are shown at the top. (Scales: 50 pA, 2 s.)(Lower) Summary of data from all experiments. Control c.c. and Control v.c.:RGCs were kept at current- or voltage-clamp mode during TBS. Lesion: Theoptic nerve was severed (n � 3) or the optic tectum was surgically removed (n �2). Colchicine: The tadpole was preincubated (1 h) in colchicine (5 �M). (B)Requirement of NMDAR activation for the enhancement of eCSCs in RGCs. Theenhancement was prevented by local bath application of AP5 (50 �M) [AP5(local)] at the optic tectum but not by loading of MK-801 (500 �M) into RGCs[MK-801 (RGC)]. (C and D) Reduction of eCSCs in RGCs induced by LFS identicalto that described in Fig. 1A. Data from an example experiment (C Upper) andsummary data for all experiments (C and D Lower) are presented in the samemanner as in A and B.

Du et al. PNAS � November 10, 2009 � vol. 106 � no. 45 � 18891

NEU

ROSC

IEN

CEIN

AU

GU

RAL

ART

ICLE

Dow

nloa

ded

by g

uest

on

Dec

embe

r 12

, 202

0

and all subsequent figures, only off responses are presented. Themean eCSC charge at 15–30 min after TBS was 130 � 9% ofcontrol (n � 6) for the off responses. Similar effects wereobserved for on responses.

Two lines of evidence argued that the potentiation effect ofTBS on eCSCs in RGCs was not caused by antidromic actionpotentials triggered by TBS, but required signals mediatedthrough the optic nerve. First, the potentiation effect was absent(96 � 7% of control, n � 5) when the optic nerve of the recordedretina was severed at the site of entry into the midbrain (at �200�m from the stimulation site) or when the optic tectum wassurgically removed before the optic nerve stimulation (Fig. 2 A).Severing of the optic nerve or ablation of the optic tectum byitself without TBS did not cause any detectable effect onlight-evoked eCSCs in the RGCs (103 � 6% of control, n � 12;Fig. S1), suggesting no immediate deleterious effect of thesesurgical operations. Second, TBS-induced enhancement ofeCSCs was absent (97 � 5%, n � 5) when the tadpole waspreincubated (for 1 h) with the microtubule-disrupting agentcolchicine (5 �M; Fig. 2A), which by itself did not affect eCSCsin the RGCs (95 � 7%, n � 13; Fig. S1). This colchicine effectsuggests the potential involvement of microtubule-based axonaltransport in mediating the retrograde signal.

The induction of LTP at retinotectal synapses is required forthe apparent retrograde potentiation at BC-RGC synapses,because blocking TBS-induced retinotectal LTP with bath-applied AP5 (Fig. 1 A) prevented the enhancement of eCSCs inRGCs (97 � 7%, n � 4; data not shown). To exclude thepossibility of direct effect of AP5 in the retina, we performedadditional local perfusion of the tectum with AP5 with a pair ofinlet and outlet pipette and monitored the perfusion with a bluedye in the perfusion solution. Under this local tectal AP5perfusion, LTP induction of retinotectal synapses was prevented(102 � 6%, n � 5), and the enhancement of eCSCs in RGCs wasalso absent (97 � 7%, n � 5) (Fig. 2B). Finally, we loaded theopen NMDAR channel blocker MK-801 (500 �M) into the RGCthrough the whole-cell recording pipette and found that TBS-induced enhancement of light responses was not affected (134 �10%, n � 5) (Fig. 2B). That intracellular loaded MK-801 hadreached the postsynaptic site of BC-RGC synapses was con-firmed by the observation that the NMDAR-mediated compo-nent of sEPSCs recorded in RGCs disappeared after MK-801loading (Fig. S2; see also refs. 31 and 32). Thus, the potentiationof BC-RGC synapses is unlikely to be attributed to potentialdirect NMDAR-dependent LTP of these synapses caused byTBS at the optic nerve head.

For all experiments described above, we have examinedchanges at BC-RGC synapses only in those RGCs exhibitingantidromic action potentials during optic nerve stimulation. In atotal of 27 antidromically activated RGCs under various exper-imental conditions, 21 exhibited potentiation of BC-RGC syn-apses after TBS. In contrast, 17 of 18 RGCs not exhibitingantidromic action potentials showed no changes at BC-RGCsynapses. Thus, TBS-induced potentiation of BC-RGC synapsesis a cell-autonomous retrograde effect and not shared by adja-cent RGCs that did not undergo LTP in their output synapses inthe optic tectum.

Depression of BC-RGC Synapses After Retinotectal LTD. Retrogrademodulation of the light-evoked responses in RGCs was alsoobserved after the induction of LTD at retinotectal synapses byapplying LFS (1 Hz, 15 min) to the optic nerve head. A reductionof the total charge associated with light-evoked eCSCs in theRGC was observed 5–10 min after the onset of LFS (the meaneCSC charge 0–15 min after the offset of LFS was 73 � 7% ofcontrol, n � 7) and lasted for �20 min (Fig. 2C), indicatingdepression of BC-RGC synapses. Similar to the TBS-inducedpotentiation, LFS-induced depression of eCSCs in RGCs was

absent when the optic nerve was severed or the optic tectum wasremoved (98 � 5%, n � 5), or when the tadpole was pretreatedwith colchicine for 1 h (96 � 5%, n � 6) (Fig. 2C). Finally, wefound that blocking LTD induction in the tectum by localperfusion of the tectum with AP5 also abolished retrogradedepression in the retina (99 � 4%, n � 5) (Fig. 2D). Further-more, intracellular loading of MK-801 into the RGC had noeffect on the depression of BC-RGC synapses (69 � 12%, n �5) (Fig. 2D), suggesting that the latter depression was not causedby potential direct induction of NMDAR-dependent LTD atBC-RGC synapses. Together with the finding on retrogradepotentiation, these results indicate that LTP/LTD induced atretinotectal synapses undergo rapid long-range retrogradespread along the optic nerve to the retina, resulting in retrogradepotentiation/depression of BC-RGC synapses.

Persistent Retrograde Potentiation/Depression Caused by SpacedLTP/LTD Induction. In the above experiments, the tectal neuronwas not recorded and could fire action potentials spontaneously.Previous studies have shown that LTP/LTD of retinotectalsynapses induced by one set of TBS/LFS was readily reversed byspontaneous spiking activity of tectal cells in vivo and that threerepetitions of TBS (with an optimal interval of 5 min) wasrequired to produce stable LTP of these synapses (33). Con-sistent with the latter report, we observed stable LTP of reti-notectal synapses by three episodes of TBS. Furthermore,recording from RGCs showed that this spaced TBS protocolresulted in a persistent enhancement of eCSCs in the RGC(136 � 7% of control, n � 6) that lasted for as long as therecording was made (up to 1 h; Fig. 3A). Similarly, three episodesof LFS (each for 10 min at 1 Hz) spaced by 5-min intervals alsoinduced a long-lasting reduction of eCSCs in RGCs (75 � 8% ofcontrol, n � 5; Fig. 3B). These persistent changes of light-evokedRGC responses may be attributed to the cumulated effects ofretrograde signals resulting from repeated TBS or LFS (33).

Bidirectional Modulation of Synaptic AMPARs on RGC Dendrites.Modifications of excitatory light responses in RGCs induced byTBS or LFS reflect changes in the efficacy of BC-RGC synapses.

Fig. 3. Persistent potentiation/depression of light-evoked excitatory re-sponses in RGCs induced by spaced stimulation protocols. (A) Persistent en-hancement of eCSCs in RGCs after three sets of TBS (spaced by 5-min intervals),each identical to that described in Fig. 1A. Example (Upper) and summary(Lower) data presented in the same manner as that in Fig. 2, with the summarydata in Fig. 2A (Control c.c.) duplicated by symbols in the light color. (Scales:50 pA, 2 s.) (B) Persistent reduction of eCSCs in RGCs after three sets of LFS(spaced by 5-min intervals), each lasted for 10 min (at 1 Hz). (Scales: 50 pA, 2 s.)Data from experiments using a single set of LFS (10 min at 1 Hz) are also shownby symbols in light color.

18892 � www.pnas.org�cgi�doi�10.1073�pnas.0910659106 Du et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 12

, 202

0

Further studies showed that spaced TBS protocol caused anincrease in the amplitude but not the frequency or time courseof spontaneous EPSCs (sEPSCs) in the RGC (Fig. S3A). Noiseanalysis of the decay phase of sEPSCs (27, 34) indicates thatspaced TBS of the optic nerve increased both the weightedsingle-channel conductance and the mean channel number ofsynaptic AMPARs (Fig. S3B), whereas spaced LFS resulted inopposite changes (Fig. S4). Thus, retrograde potentiation anddepression involves the elevation and reduction, respectively,of both the conductance and total number of postsynapticAMPARs at BC-RGC synapses.

Retinotectal LTP Depends on both Presynaptic and Postsynaptic BDNFSignaling. In Xenopus tadpoles, BDNF and its receptor TrkB areexpressed in both tectal neurons and RGCs (35, 36). We furtherexamined whether endogenous BDNF-TrkB signaling is respon-sible for the retrograde spread of LTP. Using the anti-xTrkBmorpholino oligonucleotides (MO oligos) to down-regulate thexTrkB expression unilaterally in the retina contralateral to therecorded tectal cell via injection of MO oligos into eight-cellstage embryos (Fig. S5), we found that the magnitude ofTBS-induced LTP of retinotectal synapses was significantlyreduced, as compared with that found in control tadpoles (Fig.4A, pre-TrkB KD; 134 � 9% of control, n � 6). Similarly,unilateral down-regulation of TrkB in the tectum by MO oligoinjection on the same side as the recorded tectal cell (post-TrkBKD) also resulted in a reduced magnitude of LTP induced byTBS (129 � 11%, n � 6). This result is consistent with thatobtained by inhibiting postsynaptic TrkB signaling with tyrosinekinase inhibitor k252a (200 nM) loaded into the tectal cell viawhole-cell recording pipette [k252a (TN)]. Furthermore, block-ing both presynaptic and postsynaptic TrkB activity by loading

k252a into the tectal neuron in pre-TrkB KD tadpoles resultedin complete absence of LTP after TBS (104 � 8%, n � 5).Consistently, extracellular preincubation of tadpoles with k252a(200 nM, 30 min), which would inhibit both presynaptic andpostsynaptic TrkB signaling, also prevented LTP induction (95 �6%, n � 4), whereas preincubation with the less membrane-permeable analogue k252b (200 nM) had no effect on LTPinduction (152 � 16%, n � 5; Fig. 4A).

The results of the above experiments are summarized in Fig.4B as the percentage increase in the mean EPSC amplitude at5–45 min after TBS, relative to the mean value before TBS.Furthermore, LTP induction was prevented by local infusion ofAP5 in the tectum (local AP5), but unaffected by the colchicinetreatment, which abolished the retrograde spread of LTP. Thelatter result indicates that the colchicine effect in preventingretrograde potentiation (Fig. 2 A) could not be attributed to thecolchicine effect on LTP induction. Finally, treatment with theNO synthase (NOS) inhibitor Nw-nitro-L-arginine methyl ester(L-NAME), which abolished LTD induction at these synapses(see below), had no effect on LTP induction. Together, theseresults indicate that TrkB activation by BDNF in both presyn-aptic RGCs and postsynaptic tectal neurons is required for thefull induction of retinotectal LTP (Fig. 4E), in agreement withthat found for LTP induction at CA3–CA1 synapses in thehippocampus (29).

Retrograde Potentiation Requires Presynaptic BDNF Signaling. IsBDNF-TrkB signaling also involved in the retrograde potenti-ation in RGCs after retinotectal LTP? We found that TBS-induced enhancement of eCSCs in RGCs was totally absent intadpoles that were preincubated with k252a (101 � 5% ofcontrol, n � 6), which completely blocked retinotectal LTP, but

Fig. 4. BDNF-TrkB signaling is required for retinotectal LTP and retrograde potentiation. (A) LTP induction at retinotectal synapses by TBS requires bothpresynaptic and postsynaptic BDNF-TrkB signaling. Data depict the normalized evoked EPSC amplitude of retinotectal synapses before and after TBS of the opticnerve head. Control: Standard LTP induction in control neurons, the same dataset as that shown in Fig. 1A. pre-TrkB KD: Presynaptic morpholino knockdownof TrkB in RGCs. k252a (TN): Loading of k252a (200 nM) into tectal neurons. k252a (ext): Bath application of k252a (200 nM). KD � k252a (TN): Loading of k252ainto tectal neurons in pre-TrkB KD tadpoles. (B) Summary of results from experiments shown in A, together with additional experiments performed in tadpoleswith postsynaptic morpholino knockdown of TrkB in tectal neurons (post-TrkB KD), using local application of AP5 (50 �M) at the optic tectum [AP5 (local)] orusing bath application of k252b (200 nM), L- NAME (1 mM), or colchicine (5 �M). The data represent averaged percentage increases in the EPSC amplitude att � 20–60 min after TBS relative to that of the baseline control values. Number in parentheses indicate the total number of experiments (*, P � 0.01, t test). (C)Retrograde potentiation of BC-RGC synapses after TBS requires BDNF-TrkB signaling within the RGC. Data depict the normalized eCSC charge of RGC lightresponses before and after TBS of the optic nerve head, for various treatments as those in A. Control: The same dataset as that shown in Fig. 2A (Control c.c.).k252a (RGC): Loading of k252a into the RGC. (D) Summary of all results, including those shown in Fig. 2 A and B [Lesion, Colchicine, AP5 (local), MK-801 (RGC)],together with additional experiments performed in tadpoles using postsynaptic morpholino knockdown of TrkB in tectal neurons (post-TrkB KD), or bathapplication of k252b, or L-NAME. The data represent percentage increases in the eCSC charge at t � 25–40 min after TBS (*, P � 0.01). (E) A proposed sequenceof events for the induction of LTP at retinotectal synapses (pink) and the retrograde potentiation at excitatory BC-RGC synapses (red). Note that BDNF secretedfrom presynaptic and/or postsynaptic neurons may act on both cells to generate retinotectal LTP, whereas BDNF action on the RGC may initiate the signalingresponsible for the retrograde potentiation. TN: tectal neuron.

Du et al. PNAS � November 10, 2009 � vol. 106 � no. 45 � 18893

NEU

ROSC

IEN

CEIN

AU

GU

RAL

ART

ICLE

Dow

nloa

ded

by g

uest

on

Dec

embe

r 12

, 202

0

not in those preincubated with k252b (137 � 7%, n � 5) (Fig.4 C and D). Importantly, the enhancement of eCSCs was alsototally absent when only presynaptic TrkB signaling was blockedby using either pre-TrkB morpholino KD tadpoles (101 � 5%,n � 7) or loading of k252a into the RGC of control tadpoles(102 � 7%, n � 5; Fig. 4 C and D). Note that, althoughsubstantial LTP was still present in pre-TrkB KD tadpoles, thesuppression of TrkB signaling in the RGC completely preventedthe retrograde potentiation. In contrast, suppression of postsyn-aptic TrkB activity through the use of post-TrkB KD tadpoles didnot significantly change the retrograde potentiation (125 � 7%,n � 6; Fig. 4D), suggesting that the reduced LTP generatedunder post-TrkB KD is sufficient to generate the retrogradeBDNF-TrkB signals for inducing retrograde potentiation. Asshown in Fig. 4D, the retrograde potentiation was prevented byblocking LTP induction with local AP5 perfusion, severing theoptic nerve, or pretreatment with colchicine, but unaffected byblocking LTD induction with L-NAME. Together, these resultssupport the model (Fig. 4E) that LTP induction at retinotectalsynapses by TBS is accompanied by BDNF release from eithertectal neurons or RGC axon terminals, or both, and resultantBDNF-TrkB signaling in both cells is required for the fullexpression of LTP. Furthermore, presynaptic BDNF-TrkB sig-naling in the RGC is required for the retrograde potentiation ofBC-RGC synapses.

Retinotectal LTD Requires Postsynaptic NO Signaling. Because NO isinvolved in LTD induction at several central synapses (28, 37)and Xenopus tectal neurons express neuronal NO synthase (37,38), we have examined whether endogenous NO synthesis isrequired for retinotectal LTD and the associated retrogradespread of depression to RGCs. Consistent with a previous report(28), bath application of NOS inhibitor L-NAME (1 mM)prevented LTD induction at retinotectal synapses (97 � 4% ofcontrol, n � 5) (Fig. 5 A and B). In addition, postsynaptic loadingof the membrane-impermeant NO scavenger 2-(4-carboxyphenyl)-

4,5-dihydro-4,4,5,5-tetramethyl-1H-imidazol-1-yloxy-3-oxide(cPTIO; 30 �M) into the tectal neuron (via whole-cell recordingpipette) also prevented LTD induction (97 � 5%, n � 5) (Fig. 5 Aand B). Interestingly, bath application of cPTIO (30 �M) had noeffect on LTD induction (58 � 6%, n � 7), suggesting that NOexerts its action within the cytoplasm of the tectal neuron. Fur-thermore, this NO-dependent LTD required cGMP-dependentsignaling, because LTD induction was prevented by the bathapplication of 1H-[1,2,4] oxadiazolo [4,3-a]quinoxaline-1-one(ODQ; 10 �M) (106 � 5%, n � 6), an inhibitor of NO-sensitivesoluble guanylyl cyclase (sGC) (Fig. 5 A and B). Also included inFig. 5B are results showing that LTD induction at these retinotectalsynapses was blocked by inhibiting NMDARs with local perfusionof AP5, whereas extracellular infusion with k252a (which blockedLTP induction) or pretreatment with colchicine had no effect onLTD induction. Thus, LTD induction at these synapses requiresNO/cGMP signaling in the postsynaptic tectal neuron, but notextracellular NO nor its retrograde action on the RGC axonterminals.

Retrograde Depression Requires NO Signaling. Is NO signalinginvolved in the retrograde depression induced by retinotectalLTD? We found that LFS-induced retrograde depression ofRGC light responses was prevented by either bath-appliedL-NAME (97 � 8% of control, n � 5) or intraventricular infusionof NO scavenger cPTIO near the optic tectum (97 � 7%, n � 6)(Fig. 5 C and D). Whereas depleting NO in the tectal cellcytoplasm by intracellular loading of cPTIO was required toprevented LTD induction (Fig. 5A), extracellular cPTIO wassufficient to prevent the retrograde spread of depression toBC-RGC synapses. This result is consistent with the notion thatNO synthesized in the tectal cells must diffuse through theextracellular space before its retrograde action on RGCs. Fur-thermore, the retrograde depression observed at BC-RGC syn-apses depended on the activity of protein kinase G (PKG) in theRGC, because it was largely abolished when the membrane-

Fig. 5. NO-PKG signaling is required for retinotectal LTD and retrograde depression. (A) LTD induction at retinotectal synapses by LFS requires postsynapticNO-cGMP signaling. Data depict the normalized evoked EPSC amplitude of retinotectal synapses before and after LFS of the optic nerve head. Control: StandardLTD induction in control neurons, the same dataset as that in Fig. 1B. L-NAME, cPTIO (ext), and ODQ: Bath application of L-NAME (1 mM), cPTIO (30 �M), or ODQ(10 �M). cPTIO (TN): Loading of cPTIO (30 �M) into tectal neurons. (B) Summary of all results in A, together with additional experiments using local applicationof AP5 (50 �M) at the optic tectum [AP5 (local)] or bath application of k252a or colchicine. The data represent averaged percentage decreases in the EPSCamplitude at t � 35–60 min after LFS (*, P � 0.01). (C) Retrograde depression by LFS at excitatory input synapses on RGC dendrites requires NO-PKG signalingin the RGC. Data depict the normalized eCSC charge of RGC light responses before and after LFS of the optic nerve head. Control: The same dataset as that shownin Fig. 2C (Control v.c.). L-NAME: Bath application of L-NAME. Rp-cGMPS (RGC): Loading of Rp-8-pCPT-cGMPS (5 �M) into the RGC. cPTIO (tectum): Intraventricularapplication of cPTIO. (D) Summary of all results, including those shown in Fig. 2 C and D [Lesion, Colchicine, AP5 (local), and MK-801 (RGC)], together withadditional experiments using bath application of k252a. The data represent percentage decreases in the eCSC charge at t � 30–45 min after LFS (*, P � 0.01).(E) A proposed sequence of events for the induction of LTD at retinotectal synapses (blue) and the retrograde potentiation at excitatory input synapses on RGCdendrites (green).

18894 � www.pnas.org�cgi�doi�10.1073�pnas.0910659106 Du et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 12

, 202

0

permeable PKG inhibitor Rp-cGMPS (Rp-8-pCPT-cGMPS; 5�M) was loaded into the RGC via the perforated patch record-ing pipette (91 � 8%, n � 5; Fig. 5 C and D). Fig. 5D summarizesthese findings, together with results showing that the retrogradedepression was prevented by blocking LTD induction with AP5infusion in the tectum, severing the optic nerve, and pretreat-ment with colchicine, but unaffected by blocking LTP inductionwith extracellular perfusion of k252A. Together, these resultssupport the notion that LFS triggers NO production in thepostsynaptic cytoplasm of tectal cells, where NO activates sGCand contributes to the induction and/or expression of LTD.Through diffusion to the extracellular space and subsequententry into RGC axon terminals NO may further activate PKGand its downstream effectors in RGCs, leading to the retrogradespread of depression to the BC inputs on the RGC dendrite(Fig. 5E).

DiscussionThe present study demonstrated the existence of rapid long-range retrograde spread of LTP and LTD in the intactretinotectal system, through two distinct mechanisms involvingBDNF-TrkB and NO-PKG signaling, respectively. Our find-ings showed that LTP/LTD is not localized only to the site ofinduction or neighboring synapses, but can spread retrogradelyfrom the axon terminal to the dendrite of the presynapticneuron, leading to a specific pattern of distributed synapticmodifications in the neural circuit. This finding of rapidretrograde modification adds another dimension to synapticsignaling in the polarized neuron; whereas membrane poten-tial-mediated dendrite-to-axon synaptic signaling providesrapid anterograde f low of electrical signals, cytoplasmic axon-to-dendrite retrograde signaling allows functional adjustmentsof input synapses at the dendrite based on the status of theoutput synapses at the axon terminals.

Retrograde Actions of Target-Derived Factors. More than threedecades ago, Purves (25) found that axotomy of postganglionicaxons of sympathetic ganglion neurons led to a severe depressionof preganglionic synaptic inputs and withdrawal of synapticcontacts within 3–11 days. Interestingly, exogenously appliedNGF prevented axotomy-induced synapse disintegration in sym-pathetic ganglia (26). Since then, it has been well established thattarget-derived factors can be retrogradely transported from theaxonal terminal to the soma/dendrites (22, 39, 40) and areresponsible for the regulation of neuronal differentiation andsurvival (21, 22), axon and dendrite growth (23, 24), and circuitformation (24). Our study revealed a more rapid form ofretrograde signaling between axon terminals and dendrites.Based on the delay of onset (� a few min) of retrogradepotentiation/depression at BC-RGC synapses after retinotectalLTP/LTD induction and the length of developing Xenopus RGCaxons (0.2–0.5 mm), the rate of retrograde spread of potentia-tion/depression was �1 �m/s, similar to that of fast axonaltransport, a microtubule-dependent and colchicine-sensitiveprocess. That colchicine pretreatment abolished retrograde po-tentiation and depression is consistent with the involvement offast axonal transport.

BDNF/NO Signaling and Retrograde Plasticity. Activity-induced LTPat several excitatory synapses depends on BDNF-TrkB signaling(29, 39, 41). Both BDNF and TrkB are highly expressed in theretina and the optic tectum of Xenopus tadpoles (23, 35–37).Because BDNF can be rapidly endocytosed and retrogradelytransported from axon terminals to the soma (22, 39, 40) anddendrites (40) in the form of ‘‘reactive endosomes,’’ the retro-grade spread of LTP observed here may be attributed to theaction of these endosomes on RGC dendrites, although theirpresence along the optic nerve after LTP induction remains to

be demonstrated. We have shown previously that exogenouslyapplication of BDNF in the optic tectum is sufficient to inducean immediate potentiation of retinotectal synapses and a gradualpotentiation of BC-RGC synapses, with a delay of �20 min (27).Similar to the retrograde spread of LTP shown here, BDNF-induced retrograde synaptic potentiation is also absent when theoptic nerve is severed or when the tadpole is pretreated withcolchicine. Because the TBS-induced LTP of retinotectal syn-apses also depends on BDNF-TrkB signaling in the tectum andblocking TrkB signaling in RGCs (but not in tectal neurons)prevented the retrograde spread of LTP, BDNF-TrkB signalingis the most likely mechanism for mediating the retrograde spreadof LTP. The time course of retrograde potentiation induced bytectal application of BDNF is slower (�20 min) than that foundfor the retrograde spread of LTP shown here (5–10 min). Thisdifference is likely to be caused by the slow penetration of BDNFinto the tectal tissue and gradual potentiation of retinotectalsynapses located deep within the tectum, in contrast to theimmediate LTP induction triggered by TBS of the optic nerve(Fig. 1).

Our finding that NO-PKG signaling is required for the induc-tion of retinotectal LTD is consistent with previous studiesshowing the NO dependence of LTD at several central synapses(28, 42). Expression of NOS has been shown in Xenopus tectalneurons but not in RGC axon terminals (37). We found that NOproduction in the tectal cell and the presence of extracellular NOboth are required for the retrograde depression, suggesting thatNO serves as a transsynaptic retrograde signal. However, giventhe short lifetime of NO (42), PKG or its downstream effectorsrather than NO itself are likely to serve as the long-rangecytoplasmic signal that induces the retrograde depression atRGC dendrites.

Functional Implication of Retrograde LTP/LTD. In the visual system,BDNF promotes axon growth in RGCs (35, 36), dendritearborization in cortical neurons (43), and synapse formation inthe tectum (36, 44). It is also required for the development ofocular dominance columns that requires stabilization and elim-ination of early thalamocortical connections (45). Furthermore,BDNF can be retrogradely transported from RGC axon termi-nals to the soma/dendrites (23), resulting in an enhanced growthof RGC dendrites (23) and a reduced apoptosis of RGCs (21).However, NO disrupts RGC axon growth in the optic tectum(38) and acts as a retrograde signal to increase neuronal apo-ptosis (21). Because LTP/LTD may be causally related to thestabilization/elimination of synapse, BDNF-dependent LTP andNO-dependent LTD at retinotectal synapses may contribute tothe refinement of the retinotopic map in the optic tectum (4, 37,45). Changes in excitatory synaptic inputs to RGCs have beenshown to cause dendritic remodelling (46), thus retrogradespread of LTP/LTD may lead to structural refinement of RGCdendrites (47).

Our results suggest that the retrograde potentiation anddepression in the retina accompanying the induction of LTPand LTD at retinotectal synapses may allow coordinatedstrengthening and weakening of BC-RGC synapses, respec-tively. This retrograde effect may provide a positive feedbackmechanism for activity-dependent refinement in the retino-tectal system. If the retinotectal synapses that undergo LTP orLTD represent ‘‘correct’’ or ‘‘incorrect’’ connections, respec-tively, the retrograde potentiation or depression of the BC-RGC synapses will facilitate further LTP or LTD of thoseretinotectal synapses made by the corresponding RGC, leadingto eventual stabilization or elimination of retinotectal connec-tions, respectively. Such retrograde functional coordinationbetween retinotectal and BC-RGC connections may contrib-ute to the activity-dependent refinement of retinal circuits,

Du et al. PNAS � November 10, 2009 � vol. 106 � no. 45 � 18895

NEU

ROSC

IEN

CEIN

AU

GU

RAL

ART

ICLE

Dow

nloa

ded

by g

uest

on

Dec

embe

r 12

, 202

0

e.g., stratification of BC axons and RGC dendrites in the innerplexiform layer (48).

In the developing retina, RGCs spontaneously fire rhythmicbursts of spikes that propagate across the retina, with the spikingfrequencies in the range from a few Hz to a few hundred Hz (49).Such retinal waves can induce LTP of retinogeniculate synapsesin the developing rat brain (50). The TBS and LFS used here forLTP and LTD induction may correspond to the high- andlow-frequency spontaneous spiking activities in the retina, re-spectively. Thus, physiological spontaneous activity in the retinamay cause retinotectal LTP/LTD, which in turn results inretrograde potentiation/depression of BC-RGC synapses, lead-ing to a coordinated refinement of tectal and retinal circuits.Finally, we note that in mature neural circuits this form of‘‘scaling’’ input synapses in accordance to the efficacy of outputsynapses may also serve for learning functions of neural circuitsin a manner analogous to that of back-propagation algorithmused in supervised learning of artificial neural network (30).

Materials and MethodsXenopus laevis tadpoles were staged by using the criteria of Nieuwkoop andFaber (51). For in vivo recording, stage 42–44 tadpoles were anesthetized withsaline containing 0.02% MS222 (Sigma) and incubated in Hepes-bufferedsaline containing 115 mM NaCl, 2 mM KCl, 10 mM Hepes, 3 mM CaCl2, 10mMglucose, and 1.5 mM MgCl2 (pH 7.4). For recording from tectal neurons,the skin on top of the head was removed and the brain was split open alongthe midline to expose the inner surface of the optic tectum. For recording fromRGCs, the lens was removed to expose the surface of the retina. Whole-cellperforated patch recording were made under visual control as described (27).The micropipette solution contained amphotericin B (200 �g/mL) and 115 mMK-gluconate, 4 mM NaCl, 1.5 mM MgCl2, 20 mM Hepes, and 0.5 mM EGTA (pH7.3). To activate retinotectal synapses, the axons of some RGCs were extracel-lularly stimulated at the optic nerve head. For visual stimulation, a small LCDscreen (Sony PLM-A35) was used to project uniform whole-field stimulus of 2-sduration to the tadpole’s eye. For more detail see SI Text.

ACKNOWLEDGMENTS. This work was supported by National Institutes ofHealth Grants EY01979 and NS36999 and National Basic Research of ChinaGrants 2006CB806605 and 2006CB943802.

1. Bliss TV, Collingridge GL (1993) A synaptic model of memory: Long-term potentiationin the hippocampus. Nature 361:31–39.

2. Martin SJ, Grimwood PD, Morris RG (2000) Synaptic plasticity and memory: An evalu-ation of the hypothesis. Annu Rev Neurosci 23:649–711.

3. Bi GQ, Poo MM (2001) Synaptic modification by correlated activity: Hebb’s postulaterevisited. Annu Rev Neurosci 24:139–166.

4. Katz LC, Shatz CJ (1996) Synaptic activity and the construction of cortical circuits.Science 274:1133–1138.

5. Malenka RC, Bear MF (2004) LTP and LTD: An embarrassment of riches. Neuron 44:5–21.6. Bliss TV, Lomo T (1973) Long-lasting potentiation of synaptic transmission in the

dentate area of the anaesthetized rabbit following stimulation of the perforant path.J Physiol (London) 232:331–356.

7. Andersen P, Sundberg SH, Sveen O, Wigstrom H (1977) Specific long-lasting potenti-ation of synaptic transmission in hippocampal slices. Nature 266:736–737.

8. Mulkey RM, Malenka RC (1992) Mechanisms underlying induction of homosynapticlong-term depression in area CA1 of the hippocampus. Neuron 9:967–975.

9. Lynch GS, Dunwiddie T, Gribkoff V (1977) Heterosynaptic depression: A postsynapticcorrelate of long-term potentiation. Nature 266:737–739.

10. Kullmann DM (2000) Spillover and synaptic cross-talk mediated by glutamate andGABA in the mammalian brain. Prog Brain Res 125:339–351.

11. Bonhoeffer T, Staiger V, Aertsen A (1989) Synaptic plasticity in rat hippocampal slicecultures: Local ‘‘Hebbian’’ conjugation of presynaptic and postsynaptic stimulationleads to distributed synaptic enhancement. Proc Natl Acad Sci USA 86:8113–8117.

12. Cash S, Zucker RS, Poo MM (1996) Spread of synaptic depression by presynapticcytosolic signaling. Science 272:998–1001.

13. Engert F, Bonhoeffer T (1997) Synapse specificity of long-term potentiation breaksdown at short distances. Nature 388:279–284.

14. Nishiyama M, Hong K, Mikoshiba K, Poo MM, Kato K (2000) Calcium stores regulate thepolarity and input specificity of synaptic modification. Nature 408:584–588.

15. Schuman EM, Madison DV (1994) Locally distributed synaptic potentiation in thehippocampus. Science 263:532–536.

16. Harvey CD, Svoboda K (2007) Locally dynamic synaptic learning rules in pyramidalneuron dendrites. Nature 450:1195–1200.

17. Fitzsimonds RM, Song HJ, Poo MM (1997) Propagation of activity-dependent synapticdepression in simple neural networks. Nature 388:439–448.

18. Tao HW, Zhang LI, Bi GQ, Poo M (2000) Selective presynaptic propagation of long-termpotentiation in defined neural networks. J Neurosci 20:3233–3243.

19. Ganguly K, Kiss L, Poo MM (2001) Enhancement of presynaptic neuronal excitability bycorrelated presynaptic and postsynaptic spiking. Nat Neurosci 3:1018–1026.

20. Li CY, Lu JT, Wu CP, Duan SM, Poo MM (2004) Bidirectional modification of presynapticneuronal excitability accompanying spike timing-dependent synaptic plasticity. Neu-ron 4:257–268.

21. Bahr M (2000) Live or let die: Retinal ganglion cell death and survival during develop-ment and in the lesioned adult CNS. Trends Neurosci 23:483–490.

22. Zweifel LS, Kuruvilla R, Ginty DD (2005) Functions and mechanisms of retrogradeneurotrophin signaling. Nat Rev Neurosci 6:615–625.

23. Lom B, Cogen J, Sanchez AL, Vu T, Cohen-cory S (2002) Local and target-derivedbrain-derived neurotrophic factor exert opposing effects on the dendritic arborizationof retinal ganglion cells in vivo. J Neurosci 22:7639–7649.

24. Vrieseling E, Arber S (2006) Target-induced transcriptional control of dendritic pat-terning and connectivity in motor neurons by the ETS gene Pea3. Cell 127:1439–1452.

25. Purves D (1975) Functional and structural changes in mammalian sympathetic neuronsfollowing interruption of their axons. J Physiol (London) 252:429–463.

26. Purves D (1976) Effects of nerve growth factor on synaptic depression after axotomy.Nature 260:535–536.

27. Du JL, Poo MM (2004) Rapid BDNF-induced retrograde synaptic modification in adeveloping retinotectal system. Nature 429:878–883.

28. Mu Y, Poo MM (2006) Spike timing-dependent LTP/LTD mediates visual experience-dependent plasticity in a developing retinotectal system. Neuron 50:115–125.

29. Gartner A, et al. (2006) Hippocampal long-term potentiation is supported by presyn-aptic and postsynaptic tyrosine receptor kinase B-mediated phospholipase C� signal-ing. J Neurosci 26:3496–3504.

30. Rumelhart DE, Hinton GE, Williams RJ (1986) Learning internal representations by errorpropagation. Parallel Distributed Processing, eds Rumelhart DE, McClelland JL (MITPress, Cambridge, MA), Vol 1, pp 318–362.

31. Humeau Y, Shaban H, Bissieres S, Luthi A (2003) Presynaptic induction of heterosyn-aptic associative plasticity in the mammalian brain. Nature 426:841–845.

32. Lien CC, Mu Y, Vargas-Caballero M, Poo MM (2006) Visual stimuli-induced LTD ofGABAergic synapses mediated by presynaptic NMDA receptors. Nat Neurosci 9:372–380.

33. Zhou Q, Tao HW, Poo MM (2003) Reversal and stabilization of synaptic modificationsin a developing visual system. Science 300:1953–1957.

34. Benke TA, Luthi A, Isaac JTR, Collingridge GL (1998) Modulation of AMPA receptorunitary conductance by synaptic activity. Nature 393:793–797.

35. Cohen-Cory S, Fraser SE (1995) Effects of brain-derived neurotrophic factor on opticaxon branching and remodeling in vivo. Nature 378:192–196.

36. Marshak S, Nikolakopoulou AM, Dirks R, Martens GJ, Cohen-Cory S (2007) Cell-autonomous TrkB signaling in presynaptic retinal ganglion cells mediates axon arborgrowth and synapse maturation during the establishment of retinotectal synapticconnectivity. J Neurosci 27:2444–2456.

37. Schmidt JT (2004) Activity-driven sharpening of the retinotectal projection: The searchfor retrograde synaptic signaling pathways. J Neurobiol 59:114–133.

38. Cogen J, Cohen-Cory S (2000) Nitric oxide modulates retinal ganglion cell axon arborremodeling in vivo. J Neurobiol 45:120–133.

39. Poo MM (2001) Neurotrophins as synaptic modulators. Nat Rev Neurosci 2:24–32.40. von Bartheld CS (2003) Axonal transport and neuronal transcytosis of trophic factors,

tracers, and pathogens. J Neurobiol 58:295–314.41. Korte M, et al. (1995) Hippocampal long-term potentiation is impaired in mice lacking

brain-derived neurotrophic factor. Proc Natl Acad Sci USA 92:8856–8860.42. Prast H, Philippu A (2001) Nitric oxide as modulator of neuronal function. Prog

Neurobiol 64:51–68.43. Horch HW, Katz LC (2002) BDNF release from single cells elicits local dendritic growth

in nearby neurons. Nat Neurosci 5:1177–1184.44. Alsina B, Vu T, Cohen-Cory S (2001) Visualizing synapse formation in arborizing optic

axons in vivo: Dynamics and modulation by BDNF. Nat Neurosci 4:1093–1101.45. Calbelli RJ, Shelton DL, Segal RA, Shatz CJ (1997) Blockade of endogenous ligands of

trkB inhibits formation of ocular dominance columns. Neuron 19:63–76.46. Wong WT, Faulkner-Jones BE, Sanes JR, Wong RO (2000) Rapid dendritic remodeling

in the developing retina: Dependence on neurotransmission and reciprocal regulationby Rac and Rho. J Neurosci 20:5024–5036.

47. Liu XR, et al. (2007) Brain-derived neurotrophic factor and TrkB modulate visualexperience-dependent refinement of neuronal pathways in retina. J Neurosci27:7256–7267.

48. Tian N, Copenhagen DR (2003) Visual stimulation is required for refinement of on andoff pathways in postnatal retina. Neuron 39:85–96.

49. Wong RO (1999) Retinal waves and visual system development. Annu Rev Neurosci22:29–47.

50. Butts DA, Kanold PO, Shatz CJ (2007) A burst-based ‘‘Hebbian’’ learning rule atretinogeniculate synapses links retinal waves to activity-dependent refinement. PLoSBiol 5:651–661.

51. Nieuwkoop PD, Faber J (1967) Normal Table of Xenopus laevis (North–Holland,Amsterdam).

18896 � www.pnas.org�cgi�doi�10.1073�pnas.0910659106 Du et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 12

, 202

0