connections to cerebellar cortex (larsell's hvi) in the rabbit: a...

Behavioral Neuroscience1995, Vol. 109, No. 6,1106- 1118

Copyright 1995 by the American Psychological Association, Inc.0735-7044/95/$3.00

Connections to Cerebellar Cortex (Larsell's HVI) in the Rabbit:A WGA-HRP Study with Implications for Classical

Eyeblink Conditioning

Marcy E. Rosenfield and John W. MooreUniversity of Massachusetts at Amherst

Conditioned eyeblink responses are presumably learned in the cerebellum and relayed to motoneuronsby way of the red nucleus. Projections from the red nucleus to cerebellar cortex (Larsell's lobule HVI)could be important for shaping temporally adaptive features of the conditioned response. Rabbits thathad pipettes containing wheat germ agglutinated horseradish peroxidase (WGA-HRP) implantedunilaterally into HVI showed retrograde labeling of neurons within subregions of the contralateral rednucleus implicated in eyeblink conditioning by lesioning and recording studies. Retrogradely labeledneurons were also observed in the pontine nuclei, inferior olive, and spinal trigeminal nucleus parsoralis. Projections to HVI provide a possible neural substrate for implementing time-derivativecomputational models of learning in the cerebellum. Time-derivative models are capable of describingthe timing and topography of conditioned responses.

The cerebellum is critically involved in learning and performanceof the classically conditioned eyeblink response (EBCR), a widelyused model system for neurobiological and behavioral studies ofassociative learning and memory (Thompson, 1986). There areseveral comprehensive reviews of the literature on the role of thecerebellum and associated brain stem circuits in EB conditioning(Lavond, Kim, & Thompson, 1993; Stein- metz, Lavond, lvkovich,Logan, & Thompson, 1992), as well as other varieties of adaptivemotor control (Houk, Keifer, & Barto, 1993; Thach, Goodkin, &Keating, 1992). There is also increasing interest in the cerebellumin human associative learning, including EB conditioning(Canavan, Sprengel- meyer, Deiner, H6niberg, 1994; Daum et al.,1993; Moichan, Sunderland, McIntosh, Herscovitch, & Schreurs,1994).

A series of lesion studies by Yeo and his associates (e.g.,Hardiman & Yeo, 1992; Yeo & Hardiman, 1992) and others (e.g.,Lavond & Steinmetz, 1989) indicate that a relatively small portionof cerebellar cortex, lobule HVI (Larsell, 1970), is essential fornormal expression of EBCRs in the rabbit. A recent study from thisgroup strengthened arguments that HVI is a site of learning (Gruart& Yeo, 1995). Deep cerebellar nucleus interpositus (IP) is thetarget of Purkinje cell projections from HVI. Some investigatorshave suggested that IP is a site of learning (e.g., Lavond, Kanzawa,Ivkovich, & Clark, 1994). The red nucleus (RN) of the midbrain isthe target of

Marcy E. Rosenfield and John W. Moore, Department of Psychology,Neuroscience and Behavior Program, University of Massachusetts atAmherst.

We thank Richard F. Thompson and an anonymous reviewer for theirhelpful comments. This research was supported by grants from theNational Science Foundation, Air Force Office of Scientific Research,and the University of Massachusetts.

Correspondence concerning this article should be addressed to John W.Moore, Department of Psychology, University of Massachusetts,Amherst, Massachusetts 01003-7710. Electronic mail may be sent viaInternet to [email protected].

projections from IP, another putative site of learning (seeDesmond & Moore, 1991b, for a review). Ito (1984) hasestimated that 50 deep nuclear cells project to each RN cell.Reports from several laboratories, using a variety of methods(unit recording, lesioning, inactivation), indicate that the RN isessential for EB-conditioned responding (e.g., Clark & Lavond,1993; Desmond & Moore, 1991b; Haley, Thompson, & Mad-den, 1988). The RN sends projections to motoneuron pools thatmediate the eyeblink (Rosenfield, Dovydaitis, & Moore,

1985).A recent article in this journal (Clark & Lavond, 1993)

summarizes the connections to cerebellar cortex and deep nucleiand the pathways that give rise to the conditioned andunconditioned EB responses. Their depiction (Clark & Lavond,1993; Figure 5, p. 268) incorporates most of the well- establishedanatomical findings. It shows projections from the pontine nucleithat convey conditioned stimulus (CS) information to cerebellarcortex by way of mossy fibers and unconditioned stimulus (US)information by way of climbing fibers from the inferior olivarynucleus. It also shows the projection from the RN to the abducensand facial nuclei and the inhibitory projection from IP to theinferior olive. Figure 1 is a modification of Clark and Lavond's(1993) Figure 5. It shows two additional projections to HVI ofcerebellar cortex (dashed lines) that could play an important rolein EB conditioning. One Projection is from spinal trigeminalnucleus pars oralis (SpO). The other projection is from the RN,

The SpO --@. HVI projection was first described in the cat byIkeda (1979) and Sornana, Kotchabhakdi, and Walberg (1980),using horseradish peroxidase (HRP) as a tracer. Yeo, Hardi- man,and Glickstein (1985) confirmed the existence of this projection inthe rabbit. They injected wheat germ agglutinated horseradishperoxidase (WGA-HRP) into HVI in 4 rabbits and noted bilateralretrograde labeling of sensory trigeminal neurons rostral to thelevel of the inferior olive, i.e., SpO in Olszewski's (1950)nomenclature. A subsequent study

1106

CONNECTIONS TO CEREBELLAR CORTEX

CS

UStrigeminal n. UR Abducens n.

acc.abducens n.facial n.

Figure 1. Summary of neural circuits underlying eyeblink conditioning, as modified from Clark and Lavond(1993), with dashed lines showing projections to cerebellar granule cells from spinal oralis and the red nucleus.Open circles denote excitatory units; black circles denote inhibitory units. The dotted-line projections from thered nucleus to the inferior olive (inhibitory) and spinal oralis (excitatory) are physiologically defined and havea more speculative status than the other circuit elements. CS = conditioned stimulus; US = unconditionedstimulus; UR = unconditioned response; CR = conditioned response; R = response. Copyright 1993 by theAmerican Psychological Association. Adapted with permission of the author.

1107

R

verified these findings and detailed other projections of thesensory trigeminal complex to the cerebellum, the inferior olive,and other precerebellar nuclei (Van Ham & Yeo, 1992). A single-unit recording study by Richards, Ricciardi, and Moore (1991)reported the presence of EBCR-related activity among SpOneurons. Thus, the SpO ! HVI projection provides a route bywhich information important for learning and performance, suchas the timing of the CR and the US, might ascend directly to HVIwithout first being routed through the inferior olive.

The present research was motivated primarily by the RN ! HVIprojection shown in Figure 1. Its existence was suggested byDietrichs and Walberg (1983), who injected WGA-HRP into thecerebellar cortex of 7 cats. One case involved an injection intocrus I that extended into the simplex lobe, which is homologous toHVI in the rabbit, and the rostro-lateral part of the adjacentparamedian lobe. Dietrichs and Walberg (1983) reported a total of20 retrogradely labeled neurons in the contralateral RN in thiscase, mainly in the rostral half of the structure. They characterized16 neurons as "medium size" and 4 as "large."

Following the lead of Dietrichs and Walberg (1983), we

implanted WGA-HRP-loaded pipettes into HVI of rabbits, usingthe method of Mori, Hori, and Katsuda (1981), in order confirmthe existence of the RN @, HVI projection in the rabbit. Thisprojection is of interest because RN neurons encode a variety ofconditioning-related information that could affect functioning ofHVI (Desmond & Moore, 1991b). We were especially interestedin determining the location of RN neurons that project to HVI andwhether they lie within portions of the RN that have beenimplicated in EB conditioning by recording (Desmond &Moore, 1991b) and lesion studies (Rosenfield & Moore, 1983,1985). A preliminary report has been published in abstract form.

Method

SubjectsThe subjects were 46 young adult New Zealand albino rabbits weighing

2.15 kg to 3.35 kg (M = 2.95). Most had been previously used inbehavioral experiments. Forty-five rabbits received unilateral (right)implants of WGA-HRP into the HVI portion of cerebellar cortex usingprocedures described below. Our criteria for a successful implantation wereas follows: (a) The implant was histologically

1108 ROSENFIELD AND MOORE

confirmed as being within HVI. (b) The diffusion or spread of WGA-HRP did not invade deep cerebellar nuclei, especially IP. (c) Retrogradelylabeled neurons were evident in brain stem structures known to project toHVI, based on well-established anatomical findings. Specifically, subjectswere dropped from the study if there was no labeling of the pontinenuclei, the primary source of mossy fibers, or of the inferior olivarynuclei, the source of climbing fibers. One case in five (n = 9) met all ofthese criteria. The three main obstacles to success were: (a) The tracerspread to the deep cerebellar nuclei, (b) subdural bleeding compromiseduptake and transport of the tracer, and (c) the tip of the implantationpipette would sometimes lodge within the fissure ("wishbone" structure)of HVI instead of the granule cell layer.

One rabbit received an implant into RN in order to visualize terminalsin the granule cell layer of HVI. Red nucleus implantations do not presentthe technical challenges of surgical approaches to HVI because of theabsence of porous bone and underlying vascular sinuses.

Pipette Preparation

Pipettes for implanting WGA-HRP (peroxidase-labeled lyophilizedpowder; Product number L 3892 Sigma Chemical Co., St. Louis, MO)were made from glass pipettes (AM Systems, Inc., Type 6275, Everett,Washington), Each pipette was cut into 15 mm to 30 mm lengths. Theywere tapered to an inside diameter of 25 µm to 50 µm, using a KopfModel 700 (Kopf Instruments, Tujunga, CA) pipette puller. A pipette wasthen loaded with paraffin through capillary action and air-dried to a hardwax. It was then lowered tip first into an ether bath for 10 min in order tocreate a 35 Itm to 100 Jim cavity sufficient to hold a minute quantity(approximately 0.5 mg) of dry WGA-HRP.

Surgery

The pipette containing WGA-HRP was implanted into the right HVI(or the left RN) using stereotaxic procedures. A rabbit was anesthesizedwith ketamine (Ketaset Butler, Columbus, OH 50 mg/ kg) plus xylazine(Rompun, 10 mg/kg), injected intramuscularly. Xylocaine (1%) wasinjected subcutaneously into the area of incision and the zygomaticarches. The head was positioned in a Kopf Model 900 (Kopf Instruments,Tujunga, CA) stereotaxic instrument with rabbit adapter such that lambdaand bregma were in the same horizontal plane. A scalp incision wasmade, the skull exposed, and a 4-mm hole was trephined over the targetstructure exposing the dura mater. The dura was incised, and bleedingwas controlled with thrombin (Thrombostat, Timken-Mercy MedicalSupplies, Canton, OH). The loaded pipette was lowered into the targetstructure using an electrode carrier. The area around the pipette was thenpacked with Gelfoam (Upjohn, Kalamazoo, Michigan), and the pipettewas cemented in place with dental am]ic (Teledyne-Getz, ElkgroveVillage, IL). The body of the pipette was cut flush to the cement, and theincision was cleaned and closed.

Perfusion

Fourty-four to 47.5 hr after implantation, the rabbit was anesthetized asbefore and then overdosed with 0.5 ml sodium pentobarbital (6 g/ml [0.39g/ml] Socumb, Butler, Columbus, OH), injected intravenously. The rabbitwas perfused transcardally with the descending aorta clamped. A largecannula was used to ensure a high-pressure flow of perfusates: 2 liters of0.9% saline, followed with 500-ml 10% formalin, followed with 3 litersof cold (4 °C) 12% sucrose solution.

Histology

Whole brains were removed and split into two portions. A portionconsisting of the midbrain was placed into cold 30% sucrose- phosphatebuffer until slicing. The other portion consisted of the cerebellum andcaudal brain stem. The cerebellum was embedded with gelatin andallowed to harden for 0.5 hr under refrigeration. The portion containingthe cerebellum was then put into cold sucrose- phosphate buffer for 10 to12 hr and then into cold 5% formalin for 7 hr. At this point both portionswere ready for slicing and mounting. The midbrain portion containing RNwas sliced transversely at 60 µm (frozen sections) saving every section.The portion containing the cerebellum was sliced transversely at 60 µm,mounting every section at the level of HVI and inferior olive (10) andevery other section in between. Tissue slices were placed into cold 0.1 Mphosphate buffer (pH = 7.2), mounted onto subbed slides using phosphatebuffered distilled water to prevent tissue wrinkling, and air-dried for 2 hr.Slides were then given three rinses in distilled water. They were thensubmerged for 30 min in a bath containing a mixture of two reactionsolutions. One solution consisted of sodium nitroferrocyanide (800 mg),acetate buffer (40 ml, pH = 3.3), and 740 ml distilled water. The othersolution consisted of 60 mg tetramethylbenzidine (TMB, Sigma productnumber T 2885, Sigma Chemical Co., St. Louis, MO) and 30-mi 100%ethanol. After 30 min, 0.1 ml cold 3% hydrogen peroxide was added tothe bath. The blue reaction was typically observed within 20 min. Thisstep was followed by a 30-min rinse consisting of 1,040-rnl distilledwater and 60 ml acetate buffer. The slides were then air-dried, counter-stained with neutral red (except the RN implantation case), andcoverslipped using Mount-Quick (Research Products International, Mt.Prospect, Illinois, product number 195700).

Results and Discussion

The 45 implantations of WGA-HRP into HVI yielded ninesuitable cases according to criteria stated above. These cases wereassigned numbers 1 to 9, with I representing the most rostralimplantation site and 9 the most caudal.

HVI Implantation Sites

The implantation sites and extent of spread of WGA-HRP frompipette tips within HVI are shown in Figure 2. Each row showsreconstructions of sections for each rabbit, from the most rostralto caudal extent of diffusion. The “bull's eyes” mark the locationof pipette tips. The numbers at the bottom of the figure indicatethe distance of each section in mm from an arbitrary 0-plane,defined as the level of the superior olivary nuclei (SO). Thisreference was used in histological reconstructions in previousstudies from this laboratory. In this and subsequent figures,positive sections are rostral and negative sections caudal to the 0-plane.

Figure 2 shows that most diffusion of WGA-HRP was confinedto HVI. In two cases (5 and 6) there was some diffusion intounderlying white matter at Levels 0.84 and 0.00) and theparamedian lobe (Cases 5, 6, and 8). The pipette tips were locateddeep within HVI, and their placement corresponds to the locationof recording electrodes in Berthier and Moore's (1986) study ofEBCR-related neuronal activity.

Retrogradely Labeled Cells

We defer discussion of the SpO and RN labeling results untilconsideration of the more well-established pathways, in

1


order that readers can assess the adequacy of our protocol againstthese benchmarks. The patterns of retrogradely labeled cells in thepontine and inferior olivary nuclei were criteria for establishing asuccessful implantation.

Pontine nuclei. We observed retrogradely labeled cells in thepontine nuclei for each of the successful implantations shown inFigure 2. Figure 3 shows the distribution of pontine nucleuslabeling for these nine cases. This labeling was mainly confined toLevels 2.40 through 2.20 mm. Labeled cells were evident in boththe lateral and medial portions of this structure in most cases.

Pontine nuclei project conditioned stimulus (CS) information tocerebellar cortex, as indicated in Figure 1. Wells, Hardirnan, andYeo (1989) provide a thorough review of anatomical dataregarding input to the pontine nuclei in the rabbit. Yeo et al. (1985)reported bilateral projections from the pontine nuclei to HVI.Consistent with their findings and related studies of cat (Diettichs& Walberg, 1997) and rat (Rosina & Provini, 1987), we observedmore contralateral than ipsilateral labeling of pontine nucleus cells,Referring to Figure 3, and disregarding neurons clustered along themid- line, the mean number of contralateral (left) pontine nuclearcells was 36; the mean number of ipsilateral (right) pontine nuclearcells was 16.8. This difference is statistically significant, t(8) =6.84,p < .001, 2-tailed.

Inferior olivary nuclei. Figure 4 shows reconstructed sectionsspanning Levels -3.40 through -5.76. These levels contained theinferior olivary nuclei. The dots within these structures indicate thearea of labeling but not the actual number of labeled cells, whichwere too numerous and densely packed to count with accuracy.Consistent with Yeo et al. (1985), most retrogradely labeled cellswere confined to the contralateral dorsal accessory olivary nucleus(DAO), with less extensive labeling in the medial accessoryolivary nuclei.

Figure 7E (discussed further below) is a photomicrographshowing a cluster of DAO cells from Case 4 at Level -3.40. Theseare compact multipolar neurons typical of the inferior olive inmany species, as can be seen in Figure 7F, a photomicrograph atLevel -4.50. Granules of reaction product are evident withinseveral of the cells in the focal plane.

The inferior olivary nuclei project US information to cerebellarcortex, as indicated in Figure 1, but the How of this informationcan be attenuated by the direct inhibitory projection from theinterpositus nucleus (IP) to the inferior olive indicated in thefigure. Activation of RN neurons can also inhibit olivary neuronsby way of a physiologically defined (in cat) pathway denoted inFigure I by the dotted line from the red nucleus that intercepts theinhibitory projection from IP (Weiss, 14ouk, & Gibson, 1991).

Several investigators have noted that inhibition of the inferiorolive by nucleus interpositus (or other premotor centers such asthe RN) can implement a simple delta learning rule. The deltarule states that learning occurs only to the extent that there is adiscrepancy between output of the learning device and thereinforcing signal. If the conditioned response is triggered byactivation of IP neurons, either directly or indirectly throughdisinhibition, by way of a circuit that includes the RN, then theinhibitory IP and RN projections would shut down the output ofinferior olivary cells

2

3

5

7

8

5 mm.

Figure 3. Reconstructions of transverse sections showing the distributionof retrogradely labeled pontine nuclear cells spanned by Levels 2.20 to2.40 for each of the nine successful cases. Anatomical landmarks:brachium conjunctivum (BC), lateral pontine nuclei (LPN), medialpontine nuclei (MPN).

whenever a full-blown EBCR-related activity moves through thepipeline. On these occasions, a reinforcing signal from theinferior olive would not be necessary because the conditionedresponse is well-established (e.g., McCorrnick & Thompson,

4

6

9

2.40 – 2.20

CONNECTIONS TO CEREBELLAR CORTEX 1111

Figure4. Reconstructions of transverse sections at Levels -3.40 through -5.76 showing the distribution ofretrogradely labeled cells in the inferior olivary nuclei, spinal trigerninal complex, and lateral reticular nucleusfor each of the nine successful cases. Anatomical landmarks: spinal trigeminal nucleus pars oralis (SpO);dorsal accessory olivary nucleus (DAO); lateral reticular nucleus (LRN).

1984). In fact, olivary activation of Purkinje cells by the USappears to fade out with the strengthening of the EBCR (e.g.,Moore & Berthier, 1987).

Although the inferior olive and climbing fibers are essential forconditioning, olivary neurons fire too slowly (maximum = 10 Hz)to encode information about the precise timing of the US,information that is essential for appropriate CR topography andtiming (Moore, Desmond, & Berthier, 1989). Precise timinginformation could come from other projection pathways, such asthat from SpO.

Lateral reticular nucieus. Figure4 also shows the location ofretrogradely labeled cells in the lateral reticular nucleus (LRN).These cells are known to project to ipsilateral cerebellar cortex,and so their presence simply confirms the success of theimplantations. Like Yeo et al. (1985), most LNR labeling wasconfined to the ipsilateral side of the brain stern. To date, nostudies have directly implicated the LRN in EB conditioning.

Spinal oralis. Figure 4 depicts bilateral labeling of cells in andaround SpO at the level of the inferior olive, consistent


with Yeo et al. (1985). Figure 5 shows representative sectionsrostral to this structure, ranging from Level 0.00 (Case 1) throughLevel -2.20 (Cases I to 9). Figure 7C (below) shows labeledneurons from the side of the brain stem (right) ipsilateral to theimplantation from Case 4 at Level -2.20. Thirteen labeled cellscan be seen. These cells are multipolar in form. Figure 7D showsthe homologous region of SpO on the contralateral side. Only twolabeled cells can be seen.

Although projections from SpO to HVI are bilateral, theipsilateral projection is more substantial than the contralateralprojection. Pooling the cell counts from Levels 0.00 through

-5.76, combining Figures 4 and 5, the mean number of labeledSpO neurons on the ipsilateral (right) side was 35; the meannumber of labeled cells on the contralateral (left) side was 15.6.This difference is statistically significant, t(8) = 4.38,p < .01, 2-tailed.

Richards, Ricciardi, and Moore (1991) reported that some SpOand neighboring reticular formation neurons encode informationabout EBCR topography. As discussed by Rich- ards et al.(1991), this EBCR-related activity might arise from direct orindirect projections from the RN, as indicated by the dotted linein Figure 1 (see also, Desmond &, Moore, 1991b,

Figure5. ReconstructionsoftransversesectionsatLevelsO.00through-2.20showingthedistributionof retrogradelylabeled cells in and around spinal trigeminal nucleus pars oralis (SpO) for each of the nine successful cases.Anatomical landmarks: abducens nerve (6), superior olivary nucleus (SO), facial nerve (7), lateral vestibularnucleus (LVN), medial vestibular nucleus (MVN), nucleus of the facial nerve (n7).

for relevant discussion). Richard F. Thompson (personalcommunication) has physiological evidence for a RN ! SpOprojection, consistent with a previous report in the cat (Davis &Dostrovsky, 1986), and Rosenfield & Moore (1990) observedretrogradely labeled RN neurons following implantation of WGA-HRP into the sensory trigeminal nucleus.

In addition to EBCR-related activation, most cells in theRichards et al. (1991) study responded to the unconditionedstimulus (US) and encoded the unconditioned response (UR) withhigh fidelity. The existence of EBCR-related activity and high-fidelity encoding of the UR in SpO neurons may be critical forconditioned response timing and topography, as discussed lateron.

The direct SpO ! HVI projection might explain why lesions ofHVI result in enhanced URs (e.g., Gruart & Yeo, 1995; Yeo &Hardiman, 1992). When this projection is intact, a US wouldexcite SpO neurons and this action would excite Purkinje cellsdirectly. Because Purkinje cells are inhibitory on IP neurons, theiractivation by the US would inhibit the contribution to the UR bythe pathway from the inferior olive to motoneurons by way of thelinks through IP and the RN, as shown in Figure 1. Removal ofHVI would remove Purkinje cell inhibition of IP neurons, therebyincreasing the US- triggered excitability of this circuit.

Red nucieus. The primary aim of this study was to verify theexistence of a direct projection from RN to HVI in the rabbit.Figure 6 shows reconstructions of sections spanning Levels 8.44through 7.64. This region includes rootlets of third (oculomotor)nerve. Figure 6 shows that retrogradely labeled cells wereconfined to the contralateral RN. The average number of labeledcells per rabbit was 22, with a standard error of 6. The mediannumber of labeled RN cells was 17, with a range of 10 to 63.These counts are consistent with those reported by Dietrichs andWalberg (1983) for the cat. As expected, the distribution oflabeled RN cells overlap those portions of the RN implicated inEB conditioning by recording (Desmond & Moore, 1991b) andlesion studies (Rosenfield & Moore, 1983; 1985).

Labeled RN cells are shown in Figure 7A, which is aphotomicrograph from Case 4 at Level 8.40. Seven neurons areevident. Four are multipolar and two are fusiform, both typesbeing characteristic of RN neurons. The three cells marked witharrows are also shown in Figure 7B at four times the power ofFigure 7A. Notice the granules of reaction product within each ofthe three cells.

CONNECTIONS TO CEREBELILAR CORTEX

Figure 8C is a photomicrograph (1.75x) of HVI from thisanimal at Level 2.70, the most rostral section representedin Figure 2. The arrow. marks the location of mossy fiberterminals shown under higher magnification in Figure 8D(IOx) and Figure 8E (40x). Figure 8E shows that the mossyfiber terminals form a typical glomerular complex (Ito,1984). In all, we counted 45 such terminals. Most terminalswere more rostral'than those shown in the Figure 8E. Theywere congregated in the medial (left) lobule and the base ofthe "wishbone" structure of HVI. Implantations of the tracerinto more caudal regions of the RN would be expected toyield a different distribution of terminals.

General Discussion

Our primary objective was to confirm and extend the results ofprevious fiber-tracing studies of projections to a portion ofcerebellar cortex, Larsell's HVI, involved in classical EB-conditioning in the rabbit. These projections are summarized inFigure 1. As noted above, two projections, one from spinaltrigeminal nucleus pars oralis and the other from the red nucleus,have not always been indicated in diagrams of neural circuitsinvolved in EB conditioning. What is the functional significanceof these pathways? The broader prospective of neural circuitsinvolved in EB conditioning provided by the anatomical datasuggests additional degrees of freedom in how the cerebellumperforms its function of adaptive motor control.

The revised picture of neural circuits underlying EBconditioning has implications for linking computational theoriesof learning with the cerebellum. If the cerebellar cortexparticipates in the learning of conditioned responses, then whatkind of information does the cerebellum require? What are theneural circuits by which this information reaches the crucial sitesof learning? The theoretical consensus is that learning requiresconvergent input from mossy fibers and climbing fibers. Theseideas stem from Marr and Albus (see Tyrrell & Willsbaw, 1992,for an overview). Most theorists now agree that the cerebellum iscrucial for learning the proper timing of conditioned responses,and various hypotheses have been put forward as to howcerebellar cortex adoptively adjusts the output of Purkinje cells toreflect timing information (e.g., Mauk & Ruiz, 1992; Moore &Blazis, 1989; Moore, Desmond, & Berthier, 1989).

Real-time computational models are capable of describing thetiming and topography of conditioned responses (e.g., Desmond& Moore, 1991a). Many real-time models can be reduced to abasic formulation that Sutton and Barto (1990) refer to as time-detivative (TD) models. Using Sutton and Barto's (1990) notation,Equation 1 specifies the moment-by- moment changes inassociative value, Vi, the weight of the connection between CSi,the ith of a set of potential CSs, and the US.

1113

Orthogradely Labeled Mossy Fiber Terininals

One rabbit received an implantation of WGA-HRP into the RN.Figure 8A shows the site of the pipette tip (arrow) and the radius ofspread of the tracer. This section is at level 8.76 mm, .32 mm morerostral than the most rostral section shown in Figure 6 but wellwithin the radius of diffusion of WGA-HRP. Success of thisimplantation was confirmed by retrogradely labeled neurons withinthe interpositus-dentate region of the contralateral cerebellum, asshown in Figure 8B, a photomicrograph from level -.96 mm. Itshows multipolar cells typical of deep cerebellar nuclei. Thebackground labeling in the figure are transacted fibers, not axonterminals. As with Hebbian learning rules generally, changes in associa-

(1)


Figure6. Reconstructions of transverse sections from levels 8.44 through 7.64 showing the distribution ofretrogradely labeled cells in the contralateral red nucleus (RN) for each of the nine successful cases. Anatomicallandmark: oculomotor nerve (3).

tive value are computed as the product of two factors. Thecoefficients αi and β are rate parameters. The factor Xi representsthe salience and associability of CSi, what Sutton and Barto (I990) call eligibility. Eligibility is a weighted average of previousand current strengths of CSi. The other factor, Y, representsreinforcement. Reinforcement in time-derivative models is thedifference (time derivative) between the response or output attime t, Y(t), and the response or output on the previous time step,Y(t - ∆t) (Equation 2). Any system or device that wouldimplement a time-derivative learning rule

must be capable of monitoring its output on both current andimmediately preceding time steps.

·Y = Y(t) - Y(t - Dt). (2)

Models that conform to the basic structure of Equation I (or itsequivalent differential equation) have been applied to EB-conditioning data, as reviewed by Sutton and Barto (1990). neyshow that a time-derivative model can describe the appropriatetiming and topography of conditioned responses if


Figure 7. Photomicrographs showing representative retrogradely labeled cells. A: Darkfield view of the rednucleus from Case 4 at Level 8.40 showing labeled cells (bar = 200 mm). B: Brightfield view of the rednucleus showing three labeled cells denoted in A (bar = 50 mm). C: Brightfield view of spinal oralis showinglabeled cells ipsilateral to the implantation from Case 4 at Level -2.20 (bar = 100 mm). D: Brightfield view ofspinal oralis showing labeled cells contralateral to the implantation from Case 4 at Level -2.20 (bar = 100 mm).E; Darkfield view showing labeled inferior olivary cells contralateral to the implantation from Case 4 at Level-3.40 (bar = 100 mm). F: Brightfield view of labeled dorsal accessory olivary cells contralateral to theimplantation from Case 4 at Level -4.50 (bar = 50 mm).


Figure 8. Darkfield photomicrographs illustrating terminal labeling within the granular layer of HVI followingimplantation of WGA-HRP into the red nucleus. A: Implantation site at Level 8.76 denoted by arrow (bar = 2mm). B: Retrogradely labeled cells in deep cerebellar nuclei at Level -0.96 (bar = 100 µm). C: Hemispherallobule VI at Level 2.70 (bar = .86 mm). D: Same as C showing with arrow in rectangle showing locus of threelabeled terminals in granular layer (bar = 200 µm). E: Same as D showing three labeled terminals (bar = 50µm).

one assumes the CS-US interval is segmented into a sequence oftime-tagged units, each of which develops its own associativevalue over training. The observed conditioned response is thencomputed as the summation of individual associative values overthe CS-US interval. This resembles the approach to conditionedresponse timing and topography employed by Moore and hisassociates (e.g., Desmond & Moore, 1991a; Moore et al., 1989).

We suggest that the projections to cerebellar cortexdiagrammed in Figure I provide pathways of information flowthat might implement Equation I in cerebellar cortex. First,assume that information about a CS's eligibility, X, reachescerebellar cortex by way of mossy fibers originating in thepontine nuclei (e.g., Knowlton & Thompson, 1992). The gain

of this signal, αi, is modulated by input to the pontine nuclei fromthe limbic system and neocortex (e.g., Schmajuk & DiCario,1992). Assume further that US-information reaches cerebellarcortex by way of climbing fibers originating in the inferiorolivary nuclei (e.g., Swain, Shinkman, Nordholm, & Thompson,1992). These pathways can implement a simple Hebbian learningrule, such as the delta rule mentioned previously, at the level ofPurkinje cells. However, in order to implement a time-derivativelearning rule, cerebellar cortex needs information about output,i.e., about the strength of the EB response, at two different pointsin time, t and t - ∆t. Because there are no axon collaterals fromEB motoneuron, information about output must come frompremotor sources.

The RN ! HVI projection provides a pathway by which


high-fidelity efference copy of the imminent response, Y(t), canreach Purkinje cells. Because the RN also projects to SpO, whichparticipates in the final stage of response generation, the SpO !HVI projection provides a pathway by which Y(t - At) can reachthe same place. Because of conduction and synaptic delaysbetween the RN and SpO, cerebellar cortex receives two efferencecopies of the response separated in time (i.e., therebyimplementing a time-derivative learning rule.

There are several issues raised by this interpretation of thefunctional role of the two projections in eyeblink conditioning:

1. How might TD learning be implemented in cerebellar cortex?Moore and Blazis (1989) suggested that synaptic changes in TDlearning might occur through Hebbian mechanisms operatingwithin the granular cell layer, at mossy fiber-granule cell synapses.It was suggested that synaptic changes involved the operation ofintrinsic inhibitory neurons, Golgi cells and basket cells, as well ascollateral action of Purkinje cells. Alternative schemes forimplementing TD learning in cerebellar cortex have beensuggested (Moore, Desmond, & Berthier, 1989).

2. Is there independent evidence that the RN or SpO projection isinvolved in adaptative timing of the EBCR? Two such pieces ofevidence are (a) Desmond and Moore's (1991b) and Richards etal.'s (1991) recording studies, showing that some neurons (thosedesignated as CR-locked) encode EBCR topography with highfidelty, and (b) Krupa, Weng, and Thompson's (1994) finding thatdisruption of RN function by the infusion of muscimol prevents theshift in conditioned response timing that would otherwise occurwhen the CS-US interval is increased during training.

3. When contrasted to the influence of climbing fibers onPurkinje cell activity, how can relatively sparse mossy-fiberprojections to cerebellar cortex, such as those reported here, have asignificant effect on Purkinje cell output? We suggest that theclimbing fiber activation of Purkinje cells is inhibited byinfluences from IP and the RN, discussed previously.Conditioning-related activity in these structures would attenuateclimbing-fiber activity and thereby enhance the signal-to- noiseratio of mossy-fiber inputs. In this regard, Desmond and Moore(1991b) noted the existence of CS-locked-CR-related neurons inthe RN. These cells fire a short-latency sustained burst that spansthe CS-US interval. Were they to project to 10, such cells mightinhibit climbing fiber activity for the duration of the trial, therebycontributing to the enhancement of mossy-fiber inputs to cerebellarcortex.

References

Berthier, N. E., & Moore, J. W. (1986). Cerebellar Purkinje cell activityrelated to the classically conditioned nictitating membrane response.Experimental Brain Research, 63, 341-350.

Canavan, A. 0. M., Sprengelmeyer, R., Deiner, H.-C., & Hömberg, V.(1994). Conditional associative learning is impaired in cerebellardisease in humans. Behavioral Neuroscience, 108, 475-485.

Clark, R. E., & Lavond, D. G. (1993). Reversible lesions of the rednucleus during acquisition and retention of a classically conditionedbehavior in rabbits. Behavioral Neuroscience, 107, 264-270.

Daum, I., Schugens, M. M., Ackermann, H., Lutzenberger, W., Dichgans,J., & Birbaumer, N. (1993). Classical conditioning after cerebellarlesions in humans. Behavioral Neuroscience, 107, 746-756.

Davis, K. D., & Dostrovsky, J. O. (1986). Modulatory influences of rednucleus stimulation on the somatosensory responses of cat trigemi- nalsubnucieus oralis. Experimental Neurology, 91, 80-101.

Desmond, J. E., & Moore, J. W. (1991a). Altering the synchrony ofstimulus trace processes: tests of a neural-network model. BiologicalCybernetics, 65, 161-169.

Desmond, J. E., & Moore, J. W. (1991b). Single-unit activity in red nucleusduring the classically conditioned rabbit nictitating membrane response.Neuroscience Research, 10, 260-279.

Dietrichs, E., & Walberg, F. (1983). Cerebellar cortical afferents from thered nucleus in the cat, Experimental Brain Research, 50, 353-358.

Dietrichs, E., & Walberg, F. (1987). Cerebellar nuclear afferents- where dothey originate? A re-evaluation of the projections from some lower brainstem nuclei. Anatomy and Embryology, 177, 165-172.

Gniart, A., & Yeo, C. H. (1995). Cerebellar cortex and eyeblinkconditioning: Bilateral regulation of conditioned responses. ExperimentalBrain Research 104, 431-448.

Haley, D. A., Thompson, R. F., & Madden, J. (1988). Pharmacologicalanalysis of the magnocellular red nucleus during classical conditioning ofthe rabbit nictitating membrane response. Brain Research, 454,131-139.

Hardiman, M. J., & Yeo, C. H. (1992). The effects of kainic acid lesions ofthe cerebellar cortex on the conditioned nictitating membrane response inthe rabbit. European Journal of Neuroscience, 4,966-980.

Houk, J. C., Keifer, J., & Barto, A. G. (1993). Distributed motor commandsin the limb premotor network. Trends in Neurosciences, 16,27-33.

Ikeda, M. (1979). Projections from the spinal and the principal sensorynuclei of the trigeminal nerve to the cerebellar cortex in the cat, asstudied by retrograde transport of horseradish peroxidase. Journal ofComparative Neurology, 184, 567-586.

Ito, M. (1984). The cerebellum and neural control. New York: Ravan Press.Knowlton, B. J., & Thompson, R. F. (1992). Conditioning using a cerebral

cortical stimulus is dependent on the cerebellum and brain stem circuitry.Behavioral Neuroscience, 106, 509-517.

Krupa, D. I., Weng, J., & Thompson, R. F. (1994). Involvement of the rednucleus in accurate timing of the rabbit's classically conditioned eyeblinkresponses. Society for Neuroscience Abstracts, 20, 1010.

Larsell, O. (1970). The comparative anatomy and histology of thecerebellum from monotremes through apes. Minneapolis: University ofMinnesota Press.

Lavond, D. G., Kanzawa, S. A., lvkovich, D., & Clark, R. E. (1994).Transfer of learning but not memory after unilateral cerebellar lesions inrabbits. Behavioral Neuroscience, 108, 284-293.

Lavond, D. G., Kim, J. J., & Thompson, R. F. (1993). Mammalian brainsubstrates of aversive classical conditioning. Annual Review ofPsychology, 44, 317-342.

Lavond, D. G., & Steinmetz, J. E. (1989). Acquisition of classicalconditioning without cerebellar cortex. Behavioural Brain Research,33,113-164.

Mauk, M. D., & Ruiz, B. P. (1992). Learning-dependent timing ofPavlovian eyelid responses: Differential conditioning using multipleinterstimulus intervals. Behavioral Neuroscience, 106, 666-681.

McCormick, D. A., & Thompson, R. F. (1984, January 20). Cerebellum:Essential involvement in the classically conditioned eyelid response.Science, 223, 296-299.

Moichan, S. E., Sunderland, T., McIntosh, A. R., Herscovitch, P., &Schreurs, B. O. (1994). A functional anatomical study of associativelearning in humans. Proceedings of the National Academy of ScienceUSA, 91, 81224126.

Moore, J. W., & Berthier, N. E. (1987). Purkinje cell activity and theconditioned nictitating membrane response. In M. Glickstein, C.


Yeo, & J. Stein (Eds.), Cerebellum and neuronal plasticity (pp. 338-352). New York: Plenum.

Moore, J. W., & Blazis, D. E. J. (1989). Simulation of a classicallyconditioned response: A cerebellar neural network implementation of theSutton-Barto-Desmond model. In J. H. Byme & W. O. Berry (Eds.),Neural models of plasticity (pp. 187-207). San Diego, CA: AcademicPress.

Moore, J. W., Desmond, J. E., & Berthier, N. E. (1989). Adaptively timedconditioned responses and the cerebellum: A neural network approach.Biological Cybernetics, 62, 17-28.

Mori, J., Hori, N., & Katsuda, N. (1981). A new method for application ofhorseradish peroxidase into a restricted area of the brain. Brain ResearchBulletin, 6, 19-22.

Olszewski, J. (1950). On the anatomical and functional organization of thespinal trigeminal nucleus. Joumal of Comparative Neurology, 92, 401-413.

Richards, W. G., Ricciardi, T. N., & Moore, J. W. (1991). Activity ofspinal trigeminal pars oralis and adjacent reticular formation units duringdifferential conditioning of the rabbit nictitating membrane response.Behavioural Brain Research, 44, 195-204.

Rosenfield, M. E., Dovydaitis, A., & Moore, J. W. (1985). Brachiumconjunctivum and rubrobulbar tract: Brain stem projections of rednucleus essential for the conditioned nictitating membrane response.Physiology & Behavior, 34, 751-759.

Rosenfield, M. E., & Moore, J. W. (1983). Red nucleus lesions disrupt theclassically conditioned nictitating membrane response in rabbits.Behavioural Brain Research, 10, 393-398.

Rosenfield, M. E., & Moore, J. W. (1985). Red nucleus lesions impairacquisition of the classically conditioned nictitating membrane responsebut not eye-to-eye savings or unconditioned response amplitude.Behavioural Brain Research, 17, 77-8 1.

Rosenfield, M. E., & Moore, J. W. (1990). Red nucleus projections to theaccessory abducens nucleus in rabbits reexamined with WGA- HRP.Society for Neuroscience Abstracts, 16, 271.

Rosina, A., & Provini, L. (1987). Cerebellar transcommissural neurons. InM. Glickstein, C. Yeo, & J. Stein (Eds.), Cerebellum and neuronalplasticity (pp. 63-82). New York: Plenum.

Schmajuk, N. A., & DiCario, J. (1992). Stimulus configuration, classicalconditioning, and hippocampal function. Psychological Review, 99, 268-305.

Somana, R., Kotchabhakdi, N., & Walberg, F. (1980). Cerebellar afferentsfrom the trigeminal sensory nuclei in the cat. Experimental BrainResearch, 38, 57-.64.

Steinmetz, J. E., Lavond, D. G., Ivokovich, D., Logan, C. G., & Thompson,R. F. (1992). Disruption of classical eyelid conditioning after cerebellarlesions: Damage to a memory trace system or a simple performancedeficit? Journal of Neuroscience, 12, 4403-4426.

Sutton, R. S., & Barto, A. G. (1990). Time-derivative models of Pavlovianreinforcement. In M. Gabriel & J. Moore (Eds.), Learning andcomputational neuroscience: Foundations of adaptive networks (pp. 497-537). Cambridge, MA: MIT Press.

Swain, R. A., Shinkman, P. G., Nordholm, A. F., & Thompson, R. F.(1992). Cerebellar stimulation as an unconditioned stimulus in classicalconditioning. Behavioral Neuroscience, 106, 730-750.

Thach, W. T, Goodkin, H. P., & Keating, J. G. (1992). The cerebellum andthe adaptive coordination of movement. Annual Review of Neuroscience,15, 403-442.

Thompson, R. F. (1986, August 29). The neurobiology of learning andmemory, Science, 233, 941-947.

Tyrrell, T., & Wilishaw, D. (1992). Cerebellar cortex: Its simulation andthe relevance of Marr's theory. Philosophical Transactions of the RoyalSociety, London, Series B, 336, 239-257.

Van Ham, J. J., & Yeo, C. H. (1992). Somatosensory trigeminal projectionsto the inferior olive, cerebellum, and other precerebellar nuclei in rabbits.European Journal of Neuroscience, 4, 302-317.

Weiss, C., Houk, J. C., & Gibson, A. R. (1991). Inhibition of sensoryresponses of cat inferior olive neurons produced by stimulation of rednucleus. Journal of Neurophysiology, 64, 1170-1185.

Wells, G. R., Hardiman, M. J., & Yeo, C. H. (1989). Visual projections tothe pontine nuclei in the rabbit: Orthograde and retrograde tracing studieswith WGA-HRP. Journal of Comparative Neurology, 279, 629-652.

Yeo, C. H., & Hardiman, M. J. (1992). Cerebellar cortex and eyeblinkconditioning. Experimental Brain Research, 88, 623-638

Yeo, C. H., Hardiman, M. J., & Glickstein, M. (1985). Classicalconditioning of the nictitating membrane response of the rabbit. 111.Connections of cerebellar lobule HVI. Experimental Brain Research,60,114-126.

Received January 25, 1995Revision received April 19, 1995

Accepted April 21, 1995 •

connections to cerebellar cortex (larsell's hvi) in the rabbit: a...

Documents