The majority of multisensory neurons in the cat superior colliculus (SC) are able to synthesize cross-modal cues (e.g., visual and auditory) and thereby produce responses greater than those elicited by the most effective single modality stimulus and, sometimes, greater than those predicted by the arithmetic sum of their modality-specific responses. The present study examined the role of corticotectal inputs from two cortical areas, the anterior ectosylvian sulcus (AES) and the rostral aspect of the lateral suprasylvian sulcus (rLS), in producing these response enhancements. This was accomplished by evaluating the multisensory properties of individual SC neurons during reversible deactivation of these cortices individually and in combination using cryogenic deactivation techniques. Cortical deactivation eliminated the characteristic multisensory response enhancement of nearly all SC neurons but generally had little or no effect on a neuron's modality-specific responses. Thus, the responses of SC neurons to combinations of cross-modal stimuli were now no different from those evoked by one or the other of these stimuli individually. Of the two cortical areas, AES had a much greater impact on SC multisensory integrative processes, with nearly half the SC neurons sampled dependent on it alone. In contrast, only a small number of SC neurons depended solely on rLS. However, most SC neurons exhibited dual dependencies, and their multisensory enhancement was mediated by either synergistic or redundant influences from AES and rLS. Corticotectal synergy was evident when deactivating either cortical area compromised the multisensory enhancement of an SC neuron, whereas corticotectal redundancy was evident when deactivation of both cortical areas was required to produce this effect. The results suggest that, although multisensory SC neurons can be created as a consequence of a variety of converging tectopetal afferents that are derived from a host of subcortical and cortical structures, the ability to synthesize cross-modal inputs, and thereby produce an enhanced multisensory response, requires functional inputs from the AES, the rLS, or both.
A characteristic property of many superior colliculus (SC) neurons is their ability to integrate information from different sensory modalities (Stein and Meredith 1993). “Multisensory” SC neurons receive converging visual, auditory, and somatosensory projections from numerous subcortical and cortical sources (Wallace et al. 1993; see also Edwards et al. 1979;Harting et al. 1992; Huerta and Harting 1984). Yet regardless of the specific modality convergence patterns among different multisensory SC neurons, or the species in which these neurons are found, they exhibit fundamental similarities (King and Palmer 1985; Meredith and Stein 1983,1986a; Peck 1987; Stein and Wallace 1996; Wallace et al. 1996).
Each multisensory SC neuron has multiple receptive fields, one for each modality to which it is responsive. These receptive fields are in spatial register with one another (e.g., see Stein et al. 1999). Consequently, visual and auditory stimuli originating from the same event, and hence from the same location, can fall within the receptive fields of this neuron. The responses elicited by this combination of stimuli are likely to be significantly greater than those induced by either stimulus alone and can exceed the sum of the neuron's two modality-specific responses (Meredith and Stein 1986a,b; Wallace et al. 1996, 1998). If, however, these stimuli originate from different events, and hence from different locations (i.e., the stimuli are spatially disparate), so that, for example, the visual stimulus falls within its receptive field but the auditory stimulus falls outside its receptive field, the auditory stimulus will either have no effect on the neuron's response or will depress responses to the visual stimulus; an effect believed to be mediated via direct tectopetal inhibition (Kadunce et al. 1997).
Previously it has been noted that in cat, multisensory orientation behaviors are compromised by deactivation of the anterior ectosylvian sulcus (AES) (Wilkinson et al. 1996), an extraprimary region of cortex composed of somatosensory (SIV) (Burton and Kopf 1984; Clemo and Stein 1982, 1983), auditory (FAES) (Clarey and Irvine 1986), and visual (AEV) (Benedek et al. 1988; Mucke et al. 1982;Olson and Graybiel 1987) subregions. Neurons from these AES subdivisions project heavily to the SC (McHaffie et al. 1988; Meredith and Clemo 1989; Stein et al. 1983) where they contact multisensory neurons (Wallace et al. 1993). During reversible deactivation of AES, Wilkinson et al. (1996) found that although cats were unimpaired in their ability to use modality-specific cues to locate and approach a given target (a behavior presumably involving the circuitry of the SC), their ability to use cross-modal cues to enhance their performance on this task was severely disrupted. In a brief preliminary report, Wallace and Stein (1994) showed that influences from the AES can play an important role in mediating the multisensory processes of SC neurons. This suggested that the behavioral effects of AES deactivation in Wilkinson et al. (1996) were induced via the loss of multisensory integration in the SC, a notion that is further supported by an observation that similar behavioral deficits were induced in cats with a lesion in the SC using ibotenic acid but with the AES intact (Burnett et al. 2000).
However, many cortical areas send converging projections to the deep SC (e.g., see Huerta and Harting 1984; Stein et al. 1983; Wallace et al. 1993) and can affect the properties of multisensory neurons (Clemo and Stein 1984,1986; Meredith and Clemo 1989). Thus it appeared likely that the AES might not be the only corticotectal area facilitating SC multisensory integration. Previous observations point to the rostral aspect of the lateral suprasylvian sulcus (rLS) as a likely possibility. Much like the AES, the rLS is a sensory area that sends a heavy direct projection to the SC (Stein et al. 1983) and contributes to the sensory responses of neurons in its multisensory laminae (Clemo and Stein 1984, 1986;Meredith and Clemo 1989). Thus the present study was initiated to determine whether influences from the rLS, like those from the AES, are involved in SC multisensory integration and, if so, to determine their relative contributions to this process. The results demonstrate that both cortical areas play significant roles in mediating these responses, and in some cases do so independently and in other cases cooperatively. A brief abstract of these results appeared previously (Jiang et al. 1999).
Anesthetic conditions were described earlier by Wallace and Stein (1994). Cats (n = 15) were anesthetized with ketamine hydrochloride (30 mg/kg im) and acepromazine maleate (3–5 mg/kg im). Each animal was intubated through the mouth, placed in a stereotaxic head-holder, and anesthesia was maintained with halothane (1.5–4%). An intravenous cannula was inserted for the infusion of fluids (5% dextrose Ringer, 3–6 ml/h), and the animal was then artificially respired and paralyzed (pancuronium bromide, initial dose: 0.3 mg/kg, followed by a continuous supplement of 0.1–0.2 mg/kg/h iv). Supplementary ketamine was also continuously administered (10–15 mg/kg/h iv). Respiratory rate and volume were adjusted, and the end tidal CO2 was monitored and maintained at approximately 4.0%. Body temperature was measured with a rectal thermometer and kept at 37–38°C with a circulating hot water heater. The pupil of the eye contralateral to the SC to be explored was dilated with an ophthalmic solution of 1% atropine sulfate. A contact lens was placed over the contralateral cornea to correct refractive errors and keep the cornea moist. An opaque occluding lens covered the eye ipsilateral to the SC. Craniotomies exposed the AES and rLS cortices as well as the cortex overlying the SC. The exposed cortex was covered with warm gelfoam soaked in mineral oil and a microelectrode (glass- or parylene-insulated tungsten; impedance: 1–4 M at 1 kHz) was lowered through the cortex to the SC using a hydraulic microdrive. Recording procedures were identical to those described in the following text.
All survival surgery was conducted using aseptic techniques and in accordance with the Guide for the Care and Use of Laboratory Animals (National Institutes of Health Publication 86-23) and an approved Institutional Animal Care and Use Committee protocol. Each animal (n = 2) was pretreated with a dose of ketamine hydrochloride (30 mg/kg im) and acepromazine maleate (3–5 mg/kg im) in preparation for surgical implantation. The animal was intubated, anesthesia was then maintained with halothane (0.5–3%), and temperature and fluids were maintained as described in the preceding text. The animal was then placed into a stereotaxic head-holder, and craniotomies exposed the AES, rLS, and the cortex overlying the SC. The dura was opened and the sulcal walls of the AES and the rLS were gently separated to allow cooling coils (see Fig.1) to be inserted. The area overlying the cooling coils was then packed with gelfoam, and the craniotomy was sealed with orthopedic cement.
A hollow stainless steel cylinder that provided access to the SC and that served to hold the animal's head during recording experiments (McHaffie and Stein 1983) was then attached stereotaxically to the skull over the SC craniotomy with surgical screws and orthopedic cement. Following surgery the animal was given analgesics (butorphanol tartrate, 0.1–0.4 mg/kg/6 h) as needed, and received antibiotic treatments for 7–10 days (ceftriaxone 20 mg/kg/bid or enrofloacin 5 mg/kg/bid). The initial recording session was scheduled 1–5 days after terminating the antibiotic regimen. On the day of recording, the animal was anesthetized, paralyzed, and maintained using the same anesthetic and maintenance procedures as in acute experiments. The head-holder was attached to a metal frame so that no wounds or pressure points were present during the recording session and there was unobstructed access to the body. A calibrated X-Y slide was fitted over the recording well to seal the chamber and guide the electrode to the SC. During each recording session, paralysis, and thus the immobility of the eye, was checked by repeatedly back-projecting the image of the optic disk onto a translucent plastic hemisphere, or tangent screen, using a Pantoscope. In the few cases in which ocular displacement occurred (hence the visual receptive field shifted), the neuron was either excluded or the complete series of tests was run again. At the end of the recording session, anesthetics and paralytics were discontinued, and, when stable respiration and locomotion returned, the animal was returned to its home cage.
Recording and histology
The electrode was advanced into the SC using a hydraulic microdrive. Neurons were identified by their spontaneous discharges and by their responses to sensory “search” stimuli as described in the following text, and all data were obtained from individually isolated neurons. Neuronal responses were amplified, displayed on an oscilloscope, and played through an audiomonitor. The X-Y coordinates of the electrode penetration and the recording depth of each neuron were systematically recorded. In acute experiments, electrolytic lesions (20 μA DC for 10 s) were made in each electrode penetration in which neurons were studied to aid in the histological localization of recording sites. At the end of the experiments, the animal was deeply anesthetized with barbiturate (40–55 mg/kg pentobarbital sodium) and perfused transcardially with formalin (10%). The brain containing the SC was blocked and cut into 40–50 μm frozen sections. The sections were then stained with cresyl violet.
Receptive field mapping and sensory tests
Multisensory neurons were sought using a variety of visual, somatosensory, and auditory search stimuli. The visual search stimuli were both moving and stationary flashed stimuli that were projected onto either a translucent Plexiglas hemisphere or a tangent screen. Somatosensory search stimuli consisted of mechanical taps, strokes with a camel's hair brush, manual compression of the skin, and rotation of joints. Auditory search stimuli consisted of hisses, clicks, claps, whistles, and broadband (20–20,000 Hz) noise bursts. When a neuron was encountered that proved to be responsive to any one of these sensory modalities, search stimuli from each of the other sensory modalities were also presented to determine its modality convergence pattern. The receptive field for each of the effective modalities was then mapped.
RECEPTIVE FIELD MAPPING.
Visual receptive fields were mapped manually with a Pantoscope with which bars and spots of light could be projected directly onto the hemisphere or tangent screen. The borders of each receptive field were determined by moving the visual stimulus (either a bar or spot, whichever was maximally effective) from the periphery inward from all directions until an enclosed responsive area was delimited. Auditory receptive fields were mapped using effective broadband noise bursts (generally at least 10 dB above threshold) from any of a set of 16 speakers placed 15° apart and mounted on a hoop 15 cm from the head. Azimuthal positions were first mapped with the speakers along the horizontal meridian (interaural axis). The hoop was then rotated around the interaural axis to map the elevation component of each receptive field (see Meredith and Stein 1986a,b). Somatosensory receptive fields were mapped in a similar fashion, using minimally effective stimuli to avoid spuriously large receptive fields as a consequence of mechanical transmission of stimuli across the skin or subcutaneous tissue. Each receptive field was plotted on standardized representations of visual, auditory, and somatosensory space (seeStein and Meredith 1993) and reexamined during different stages of the recording series (i.e., during and after cortical deactivation).
Once a multisensory neuron's receptive fields were mapped, its responses to sensory stimuli were studied quantitatively with the stimuli described in the following sections. The onset, duration, and physical parameters of the stimuli, as well as the cross-modal onset asynchronies and intertrial intervals, were controlled independently.
Visual stimuli consisted of computer-controlled moving bars of light generated by a Barcodata or Prado projector. Bars and spots of light (0.11–13.0 cd/m2 against a background of 0.10 cd/m2) were projected onto the hemisphere or tangent screen and could be moved in all directions across the receptive field at amplitudes of 1–110° and speeds of 1–400°/s.
Somatosensory stimuli were computer controlled and consisted of indentations of the hair and skin. A probe with a tip of variable size was attached to a modified moving-coil vibrator (Ling 102A shaker) via a horizontal arm that increased the amplitude of its vertical excursion. The tip of the probe was loaded against the hair or skin and stimulation consisted of an initial displacement, a plateau phase, and a return to the “rest” position. The amplitude of the excursion could vary from 0.05 to 5.0 mm, and stimuli could be presented at velocities of 15–420 mm/s (Clemo and Stein 1986).
Auditory stimuli were computer-controlled broadband noise bursts delivered from any one of the hoop-mounted speakers. The duration of the auditory stimuli varied from 20–150 ms at intensities of 52–70 dB SPL against a background SPL of 51.4–52.0 dB.
MULTISENSORY NEURONS AND MULTISENSORY TEST PARADIGMS.
A multisensory neuron is defined as one that responds to stimuli from more than one modality or whose responses to one modality are altered by the presence of a stimulus from a second modality (Meredith and Stein 1983). Each multisensory SC neuron's response profile was determined using the same paradigm: responses to each modality-specific stimulus (e.g., visual alone, auditory alone) and to the multisensory combination (e.g., visual-auditory) were determined quantitatively by presenting the stimulus and stimulus combination in the relevant receptive fields 8 to 12 times at 8 to 20 s intertrial intervals with the different conditions interleaved. Generally, during multisensory trials the two sensory stimuli were presented simultaneously or within 20–200 ms of one another. The specific interval and stimulus parameters (e.g., intensity, size, position, motion direction, and speed, etc.) of each modality-specific stimulus that produced the greatest multisensory response enhancement were determined for each neuron during an initial set of trials that preceded actual data acquisition. Because previous studies have demonstrated that multisensory response enhancements are maximal when the individual modality-specific stimulus components of a cross-modal stimulus combination are poorly effective (see Meredith and Stein 1986a), the individual modality-specific stimuli were often minimally effective.
In acute experiments, a 5 mm2 hollow metal probe was placed directly on the dura overlying the cortical area of interest. The probe tip was continuously circulated with a refrigerated solution of water and methyl alcohol. This produced a cortical temperature of 10–12°C within 1 mm of the probe tip and a gradient of increasing temperature at progressively greater distances (Clemo and Stein 1986). The temperature of a cortical area approximately 3 mm around the probe tip was sufficiently low to produce a reversible block of cortical neurons and could be repeated many times throughout an experiment (see Clemo and Stein 1986; Ogasawara et al. 1984 for more detailed discussions of this technique).
In chronic experiments, cortical deactivation was accomplished via indwelling cooling coils placed between the sulcal walls. With this technique, the influences of the areas of interest (i.e., AES and rLS) could be examined more fully by targeting the two cortical areas simultaneously as well as sequentially. In addition, the cortex could be actively rewarmed through the same coils. By using these techniques in multiple experiments over several months, the corticotectal influences of these areas could be examined in a large population of SC neurons in the same animal. The cooling coils (7–11 × 3–7 mm) were prefabricated by fashioning loops of 19 or 21 gauge stainless steel hypodermic tubing (see Fig. 1, also see Horel 1991; Lomber et al. 1999) that were shaped appropriately for the sulci. Refrigerated water (0°C) was circulated through the coils to deactivate the cortex, and warm (36–38°C) water was circulated through them to reactivate cortex. Two coils were required to span AES, and one was placed in the rostral pole of rLS (see Fig. 1). The stems of the coils were fixed to the skull with surgical screws and orthopedic cement. The same procedure was used in two acute animals in which the temperature gradients generated by the coils were measured using a BAT-10 microprobe thermometer (Physitemp Instr.). For this purpose, an electrode was attached to the microprobe thermometer, the array was placed in a micromanipulator, and the tips of the thermoprobe and microelectrode were lowered approximately 1.5 mm below the surface of the cortical mantle at various distances from the coil. At each measurement site, the baseline temperature was recorded, then the coolant was circulated through the coils implanted in AES, rLS, or both, and the temperature changes were recorded over time with a minimal interval of 5 s. Spontaneous and sensory evoked neuronal activity was displayed on the oscilloscope, and the temperatures at which this activity changed were noted. Once the temperature decrement at a given site and for a given condition (e.g., AES deactivation) was stabilized, rewarming began and the assessment of cortical temperature was recorded in the same manner. The entire procedure was then repeated at that site for a second condition (e.g., AES/rLS deactivation). The array was removed, relocated to another cortical site (or the surface of the SC), and the procedure was begun again.
To assess the integrity of the cortical areas with cryogenic coil implants, 2% wheatgerm-agglutinated horseradish peroxidase (WGA-HRP) was injected stereotaxically into the SC of the two chronic experimental animals at the site at which the highest concentration of recordings was made. A normal control received the same treatment. The animals were perfused 36–40 h later with 1% paraformaldehyde. A block of cortex containing the AES, rLS, and the brain stem was removed from each animal, and separate sets of sections were chosen for HRP reaction (tetramethylbenzidine methods) and cresyl violet staining. Additional sets of brain stem sections containing the SC in each animal were reacted with 3′, 3′-diaminobenzidine (DAB) and stained with neutral red to localize the injection site. Although gross examination revealed that there were some physical changes in the cortical tissue due to the mechanical displacement by the cooling coils, the density and distribution of retrogradely labeled HRP neurons in both AES and rLS of the experimental animals were indistinguishable from that found in the normal control animal. The distribution of retrogradely labeled neurons in AES and rLS was also consistent with previous observations (Stein et al. 1983), and the absence of substantial cortical damage induced by cooling coil implantation is in agreement with previous studies using this cortical deactivation technique. These issues are dealt with in greater detail by Lomber et al. (1999).
Data acquisition and analysis
Each neuron served as its own control in assessing the effects of cortical deactivation. The identical stimuli and interstimulus intervals were employed during the baseline, or predeactivation “control,” condition and during cortical deactivation trials. For each neuron (time permitting), first one cortical area alone (e.g., AES), then the other area alone (e.g., rLS), and finally both areas together (AES and rLS) were inactivated. The first area deactivated alternated from one neuron to the next to control for any possible aftereffects of the preceding cortical deactivation/reactivation procedure and to ensure that a reasonable sample of SC neurons would be examined in each cortical deactivation condition (not every neuron was likely to be held long enough to complete all three cortical deactivation conditions). The duration of deactivation of a cortical area was about 10–35 min depending on the complexity of a given test series. Thus for example, after testing the effects of AES deactivation on SC responses and then reactivating that cortex, a second control series was conducted. The second cortical area (rLS alone) was then deactivated, and the entire series of tests was repeated. The testing was then repeated a third time during which both AES and rLS were deactivated and subsequently reactivated. To avoid misinterpreting response changes due to possible mechanical factors, neurons were excluded from consideration if they showed a significant response reduction during cortical deactivation (e.g., loss of multisensory integration) but did not have their characteristic pretest response enhancement reinstated after cortical reactivation. Because multisensory response enhancement has been found to be the most reliable index of multisensory integration [i.e., it is found in all SC neurons capable of integrating cross-modal cues, whereas multisensory response depression is found in only a select proportion of these neurons (see Kadunce et al. 1997)], it was used here to assess the influence of cortex on multisensory SC responses.
Each neuron's responses (number of impulses) to each modality-specific and cross-modal stimulus combination were measured using the time window that bracketed the longest response train. Although spontaneous rates were often extremely low, the stimulus-evoked responses were corrected by subtracting spontaneous activity (i.e., the number of spikes measured in the 1-s interval preceding the onset of the first stimulus and normalized for the time window in which responses were counted). Statistical analysis was performed with SYSTAT (SPSS). The criterion for multisensory response enhancement was the same under all conditions: a statistically significant (Student's t-test,P < 0.05) increase in the response to the cross-modal stimulus combination as compared with the dominant modality-specific response. Based on this criterion, multisensory neurons were segregated into two groups, those that integrated multisensory cues (i.e., exhibited multisensory response enhancement) and those that did not.
In instances in which the statistical criterion for multisensory response enhancement was obtained in a given neuron, the magnitude of the enhanced multisensory response was calculated using the following formula where CM = the mean number of impulses evoked by the cross-modal stimulus and SMmax = the mean number of impulses evoked by the “dominant” modality-specific stimulus.
For each neuron, the effect of cortical deactivation (i.e., treatment) on the modality-specific responses was determined by a Student'st-test. The difference in the multisensory response generated in the control and cortical deactivation conditions was assessed using ANOVA (a possible interaction between the treatment and the modality-specific response was incorporated into the analysis).
To explore the functional relationship between the multisensory and the modality-specific responses at the population level, each neuron's multisensory response was plotted against its own dominant modality-specific response in one graph as well as the sum of its two modality-specific responses in another graph. In each, a linear relationship was determined by a significant Pearson correlation coefficient (P < 0.05), and the equation was computed with a regression analysis (least squares). Systematic within-group (e.g., integrative neurons) differences in the magnitude of the multisensory versus the modality-specific responses at the population level were determined using repeated measures ANOVA (ANOVArm). ANOVArm was also used to compare possible differences in population in modality-specific responses induced by cortical deactivation.
Between-group analysis at the population level was conducted using an analysis of covariance (ANCOVA) linear procedure. This evaluated potential differences between two different neuronal groups (e.g., integrative and nointegrative) with respect to their relationships in multisensory versus modality-specific responses. The same analysis was used to compare the effect of cortical deactivation of each population. In these analyses, the multisensory responses were used as the dependent measure and the modality-specific responses as the covariate. Possible interactions between neuronal groups (i.e., integrating vs. nonintegrating), or treatments (i.e., cortical deactivation vs. control), and modality-specific responses were incorporated in the ANCOVA model.
Effects of chronic cryoblockade
To determine the time course of temperature changes induced by the indwelling cooling coils, and their effects on cortical activity, the cortical temperature gradients in AES and rLS were measured at several distances from each coil while recording from individual cortical neurons (see methods). The temperature gradients in the two areas were very similar and the data are combined in Fig. 1. The temperature in the cortex adjacent to a cooling coil declined steeply during the first minutes of cooling, slowed during the second minute, and then reached a stable level soon thereafter. An inverted time course was seen during cortical reactivation (rewarming). Although the individual AES and rLS neurons sampled varied in their thresholds for deactivation, virtually all neurons ceased their activity when the temperature fell to 16°C (see Fig. 1, C and D,for examples). Based on these data, the maximal effective range of blockade was estimated to be within a 2 mm radius around the 21-gauge coil (as shown in Fig. 1) and about 3 mm around the larger coil (19 gauge). These temperature gradients and cortical deactivation thresholds were similar to those noted previously in other studies using cryogenic probes of different types (e.g., Clemo and Stein 1984; Ogasawara et al. 1984) and sizes (Horel 1991; Lomber et al. 1999). In cases in which both AES and rLS coils were cooled simultaneously, the temperature decrement in the tissue directly around each coil (less than 1.5 mm) showed only minor changes (1–2°C) but was greater by approximately 4°C on the intervening gyrus. Thus at the narrowest mid-point between the anterior AES and the rLS coils (approximately 2.5 mm from each), the temperature reached a minimum of 21°C when one coil was cooled and 17°C when both coils were cooled simultaneously.
A total of 128 multisensory neurons in the SC were studied. The general response properties, topographic organization, and modality convergence patterns of these neurons, as well as the cross-modal register among their receptive fields, closely matched previous descriptions (see, for example, Meredith and Stein 1996;Stein et al. 1976; Wallace and Stein 1997). The multisensory responses of 90 of these neurons were examined quantitatively and tested during cortical deactivation. In 70 of them, cross-modal stimulus combinations evoked significantly greater responses than did their dominant modality-specific stimulus, thereby meeting the criterion for “multisensory response enhancement” (Meredith and Stein 1983). These neurons were operationally designated “integrative” multisensory neurons. In contrast, 20 of the neurons showed no significant differences between their responses to the cross-modal combination and their responses to their dominant modality-specific stimulus. These were operationally designated “nonintegrative” multisensory neurons. These proportions of integrating and nonintegrating multisensory neurons are consistent with previous reports (see Kadunce et al. 1997; Wallace and Stein 1997).
Multisensory enhancement and cortical deactivation
The response characteristics of the populations of integrative and nonintegrative multisensory neurons are depicted in Fig.2 A, where the mean multisensory response of each neuron was plotted against its dominant modality-specific response. As demonstrated by the results of linear regression analysis, the magnitude of the multisensory response of both integrating and nonintegrating neuronal populations covaried significantly with their dominant modality-specific responses (P < 0.001). However, in only the integrating neuronal group was the multisensory response enhanced significantly [ANOVArm: F(1,69) = 129.4,P < 0.001] above the dominant modality-specific responses (i.e., unity). Similar effects were apparent when the multisensory responses were plotted against the sum of their modality-specific responses as shown in Fig. 2 B. Only the integrative neuronal group had its multisensory response significantly enhanced above the sum of its modality-specific responses [ANOVArm: F(1,69) = 55.2,P < 0.001].
The apparent difference between the multisensory response profiles of the integrative and nonintegrative neuron groups was further studied by ANCOVA analysis. This revealed a statistically significant interaction between the neuronal group and the dominant modality-specific response [F(1,88) = 5.07, P < 0.05, Fig.2 A] as well as between the neuronal group and the sum of modality-specific responses [F(1,88) = 7.19,P < 0.01, Fig. 2 B], suggesting a significant difference in the slopes of the two regression lines (integrating: solid lines; nonintegrating: dashed lines). The overall gain (determined by both slope and offset) in the cross-modal responses of the integrative neurons was significantly higher than that of their nonintegrative counterparts at comparable levels of dominant modality-specific [ANCOVA: F(1,88) = 13.26,P < 0.001, Fig. 2 A] and sum of modality-specific responses [F(1,88) = 10.748,P < 0.01, Fig. 2 B].
It is worth noting that when the multisensory response enhancement shown in Fig. 2 A is normalized and dealt with as a proportionate change relative to the dominant modality-specific responses, an inverse relationship becomes apparent. Thus the proportionate response enhancement was greater at the lower range of the dominant modality-specific response, a finding in accord with the principle of inverse effectiveness (Meredith and Stein 1986a). For example, the formula predicts that for any given neuron with a dominant modality-specific response of 1, the multisensory response would be 4.6, thereby resulting in an enhancement of 360%, but for a neuron in which the modality-specific response is 15, the multisensory response would be 24.2, yielding an enhancement of only 64%.
Fifty-nine of the integrative neurons were maintained long enough to quantitatively compare the relative influence of independently deactivating AES and rLS on their modality-specific and multisensory responses. Although in all but two of these cases the neurons continued to respond to both modality-specific stimuli during cortical deactivation, most of them exhibited a striking loss of their capacity to integrate these stimuli when they were presented in combination. In such circumstances, the neuron's responses to the combination of cross-modal stimuli were generally rendered no different from its responses to one or the other of the modality-specific stimuli. Deactivation of AES produced this effect in more of these neurons (48/59: 24 were affected by AES only and 24 by independent deactivation of AES or rLS, see Table 1) than did deactivation of rLS (26/59: 2 affected by rLS only and 24 by AES or rLS, Table 1). However, the sample of neurons obtained was nearly equally divided among those whose multisensory integration capabilities depended on the integrity of only one of these cortical areas (n = 26) and those that were influenced by both areas (n = 24), thereby exhibiting “dual-dependencies.” These differing populations of SC multisensory neurons are dealt with separately in the following text.
SC neurons in which multisensory integration was dependent only on influences from AES
Multisensory response enhancement was compromised during AES deactivation but unaffected by rLS deactivation in 41% (24/59) of the SC neurons examined. Figure 3 provides a characteristic example of one such neuron. Although neither stimulus was highly effective alone, their combined presentation produced a response enhancement that was on average nearly one-and-a-half times the magnitude of the dominant modality-specific response (i.e., auditory, Fig. 3 A). Deactivation of AES resulted in a reduction of this multisensory response so that it was now statistically indistinguishable from the dominant modality-specific response, hence, multisensory response enhancement was eliminated (compare Fig. 3, B and A). Nevertheless, the neuron retained its “multisensory” identity: it continued to respond to each of the modality-specific stimuli, and its responses to these stimuli were not significantly altered during AES deactivation (t-test, P > 0.1). Reactivation of the AES reinstated the enhanced response to the cross-modal stimulus combination (Fig. 3 C).
The absence of a statistically significant effect of AES deactivation on the vigor of the modality-specific responses of multisensory SC neurons was quite common. Of the 59 neurons tested during AES deactivation (including the synergistic neuron group, see following text), only 18.6% (n = 11) of them showed statistically significant (t-test, P < 0.05) changes in their modality-specific responses during cortical deactivation (but see Clemo and Stein 1986;Meredith and Clemo 1989, where some auditory and somatosensory effects were noted when modality-specific SC neurons were included in the sample). Yet even in these cases, an increase or a decrease in the modality-specific response was not a reliable predictor of the effect that cortical deactivation would have on the neuron's multisensory response. Multisensory response enhancement was lost in some neurons and retained in others, regardless of whether the magnitude of their modality-specific response increased (n = 5/48) or decreased (n = 6/48) during AES deactivation. Deactivation of AES also produced significant alterations in the modality-specific responses of 4/20 neurons (2 increased, 2 decreased) that were incapable of multisensory integration.
SC neurons in which multisensory integration was dependent only on influences from rLS
In some cases, the capacity of SC neurons to exhibit enhanced multisensory responses depended on influences from rLS rather than AES. In these instances, the selective effects of rLS deactivation on an SC neuron's multisensory responses were indistinguishable from those induced in other neurons by the deactivation of AES. However, the incidence of such neurons was far lower. In only 2 of the 59 (3.4%) neurons examined did rLS, but not AES, deactivation affect SC multisensory integration. The results obtained from one of these neurons are illustrated in Fig.4. Note that both the visual and auditory responses were weak in this neuron, but their combination produced a response enhancement that was more than threefold the magnitude of the dominant modality-specific response (Fig. 4 A). Deactivation of rLS eliminated the enhanced multisensory response despite producing a significant (t-test, P < 0.05) augmentation of the visual response (compare Fig. 4, B and A). In contrast, deactivation of the AES significantly (P < 0.05) augmented both modality-specific responses but did not eliminate multisensory response enhancement (Fig. 4 D). Although there was a decrease in the percentage of the multisensory response enhancement during AES deactivation, it was simply due to the increased magnitude of the dominant modality-specific response (thus the magnitude of the multisensory response was very similar to the control, compare Fig. 4, D and C). Combining AES and rLS deactivation did not produce a significantly greater degradation of multisensory response enhancement than did rLS deactivation alone (compare Fig. 4, F and B, ANOVA,P > 0.1). As was the case during AES deactivation, only rarely did rLS deactivation affect the magnitude of the modality-specific responses of multisensory neurons (4/59 integrative neurons; 2/16 nonintegrative neurons).
SC neurons in which multisensory integration was dependent on converging influences from AES and rLS
Whereas the multisensory integration capabilities of the SC neurons described in the preceding text depended exclusively on influences from either AES or rLS, in many cases the integrity of both cortical areas had to be maintained for a neuron in the SC to express this capacity. Such a dual corticotectal dependency, or corticotectal synergism, was noted in 24/59 neurons (Table 1) and is typified by the example presented in Fig. 5. Although each of the modality-specific stimuli reliably activated this visual-somatosensory neuron, their combined presentation produced the characteristic multisensory enhanced response; a response that was more than threefold the magnitude of the response to the dominant modality-specific stimulus (visual, Fig. 5 A). Deactivation of either rLS or AES had the same effect: the elimination of the multisensory response enhancement (see Fig. 5, B andD). Although the modality-specific responses of this particular neuron were degraded during cortical deactivation (t-test, P < 0.05), this occurred in a minority of these neurons (16%, n = 4/24, 2 increased and 2 decreased), and the sign of the change was not predictive of the effects of cortical deactivation on their multisensory responses.
The observation that both AES and rLS were required to function simultaneously in order for many SC neurons to exhibit multisensory integration (corticotectal synergism) raised the possibility that the converse might also be true: a population of SC neurons might exist that would be capable of maintaining multisensory response enhancement as long as either one of the two cortical areas is operational (corticotectal redundancy). This class of neurons would not have been evident from the tests described in the preceding text in which the combined deactivation of AES and rLS was not conducted routinely. Indeed, to provide an accurate estimate of the incidence of the various possible corticotectal dependencies among integrative SC neurons, it was necessary to examine the multisensory responses of each SC neuron during deactivation of each cortical area independently as well as during their combined deactivation, a sequence involving seven interleaved series of cortical deactivation controls and tests. A small subpopulation of 27 neurons was maintained for the necessary period to run a complete set of these test and control trials. The results obtained indicate that nearly all (25/27, 93%) multisensory SC neurons depended on cortex for the expression of multisensory response enhancement. Interestingly, when the results of analysis of these 27 neurons were compared with the data from the larger population (see Table 1), there was good agreement: approximately 41% of the SC neurons depended solely on AES, very few depended on rLS alone (4%), and the majority (52%) depended on dual inputs from both cortical areas (Fig. 6). Moreover the results obtained from the subset of 27 neurons revealed that a category of SC neurons with redundant corticotectal dependencies did exist. Approximately 11% of the neurons sampled were able to maintain their multisensory response enhancement capability as long as either AES or rLS remained functional.
An example in which AES-rLS redundancy was evident is provided in Fig.7. During control trials, the dominant modality-specific responses of this neuron were enhanced by nearly fourfold by the multisensory stimulus combination (Fig. 7 A); this response enhancement was maintained during the selective deactivation of AES (Fig. 7 B) or rLS (Fig.7 D) and was eliminated only when both cortical areas were deactivated simultaneously (Fig. 7 F). Similar to the neurons described in the preceding text, the loss of multisensory enhancement was not associated with any significant alteration in the neuron's modality-specific responses (compare Fig. 7, F andE, t-test, P > 0.1).
Specificity of corticotectal influences on multisensory responses
The specificity of AES/rLS corticotectal influences on SC multisensory responses described in the preceding text was particularly evident in a pair of neurons whose dominant modality-specific and multisensory responses were systematically altered. This was accomplished by varying the physical parameters (e.g., size, intensity) of one of the modality-specific stimuli to produce progressively larger responses. The presentation of these stimuli, in turn, was interleaved with an invariant, or “standard,” stimulus from the other modality, and the trials were run in control and cortical deactivation conditions. This produced parallel sets of graded modality-specific and multisensory responses: one in the presence and the other in the absence of corticotectal influences. Both neurons showed similar results, and the data from one of them are illustrated in Fig.8. This visual-auditory neuron responded with progressively greater modality-specific (i.e., visual) and multisensory responses to progressive increases in the intensity of the visual stimulus. At each step increase in the intensity of the visual stimulus, the multisensory response exceeded the visual response. This was most evident at the highest intensities, where the neuron's visual responses appeared to have reached a plateau, but an even greater response could be produced when the visual and auditory stimuli were combined. Thus the modality-specific and multisensory response curves never converged. Whether or not this proves to be a general rule and whether or not the “channel capacity” for sensory processing can be reached in multisensory SC neurons only with multisensory stimuli requires a far more extensive data set.
Simultaneous deactivation of AES and rLS eliminated the enhanced multisensory responses at each stimulus level without appreciably influencing modality-specific responses. This resulted in the plot of multisensory responses appearing to collapse onto the plot of modality-specific responses (see Fig. 8, bottom right).
The selective impact of deactivating AES and/or rLS on the multisensory response enhancement of SC neurons was also underscored by the results of population analyses. These were accomplished by pooling the results obtained from all neurons whose multisensory response enhancements were affected by cortical deactivation (n = 53). The magnitude of each neuron's multisensory response was plotted against the magnitude of its dominant modality-specific response before (open circles) and during (filled circles) cortical deactivation (Fig.9). A similar plot was made of the multisensory responses of these neurons against the sum of each neuron's modality-specific responses (Fig. 9, inset). Cortical deactivation produced no consistent changes at the population level in either the dominant modality-specific response [ANOVArm: F(1,52) = 0.358,P > 0.1] or the sum of the modality-specific responses of these neurons [ANOVArm:F(1,52) = 1.714, P > 0.1].
As can be seen in Fig. 9, cortical deactivation induced a systematic and significant downward shift of the multisensory responses of integrative neurons [ANOVArm: F(1, 52) = 60.7, P < 0.0001] and a significant decrease in the slope of the regression lines [ANCOVA: F(1, 104) = 4.1, P < 0.05]. The response profiles of these neurons now became indistinguishable from those of their dominant modality-specific responses (the 2 lines now overlap) and slightly below, but not significantly different from, the sum of their modality-specific responses (Fig. 9, inset). Consequently the multisensory response of the integrative neuron group was rendered equivalent to that of the nonintegrative neuronal group (compare filled circles in Fig. 9 and empty triangles in Fig. 2), while the response profiles of the nonintegrative neurons were not affected by cortical deactivation (Fig. 10).
The present results demonstrate that the characteristic ability of SC neurons to enhance their sensory responses by synthesizing cross-modal inputs is dependent on cortical influences involving two adjacent cortical areas: the AES and the rLS. Deactivating these cortical areas eliminated the synthesized multisensory responses of nearly all SC neurons and rendered their multisensory responses indistinguishable from those of their nonintegrative counterparts. Presumably, the normal response differences between these two SC neuronal populations reflect differences in their access to AES- and/or rLS-related corticotectal projections.
Overall AES appeared to be a more important mediator of SC multisensory integration than was rLS, as a substantially larger proportion of neurons depended solely on AES than on rLS for this capability. However, it is possible that the apparent dominance of AES is due to an underestimate of the relative importance of the rLS. Given that there is no clear demarcation of the functional borders of rLS, the area deactivated in the present experiments may have been incomplete. Nevertheless it is interesting to note that the proportion of multisensory SC neurons previously found to be unresponsive to AES activation, and thus presumably lacking AES inputs (30%, see Table 1) (Wallace et al. 1993) is similar to the sum of nonintegrative multisensory SC neurons plus those integrative neurons not affected by AES deactivation in the present sample (22 + 11%). The preeminent role for AES in SC multisensory integration is also consistent with the striking deficits in SC-mediated multisensory orientation/approach behaviors that are incurred by its selective deactivation (Wilkinson et al. 1996). Yet despite a paucity of SC neurons whose multisensory integration capability depended solely on rLS, about half of the population of SC integrative neurons depended on a synergy between AES and rLS, and still others received redundant inputs from these areas and could, therefore retain their multisensory integrative capability as long as either AES or rLS was functional. The importance of corticotectal influences from outside the AES for SC multisensory integration suggests that the selective loss of these influences might also have a detrimental effect on attentive and orientation responses. Preliminary observations are consistent with this possibility (Jiang et al. 2000).
The most direct route by which AES and rLS influences can be exerted on SC multisensory integration is via their heavy direct corticotectal projections (see Stein et al. 1983), but the presence of reciprocal connections between these two cortical areas (Miceli et al. 1985; Scannell et al. 1995) makes it possible that, at least in the case of synergy, AES-rLS intracortical connections may also play a role. Given the presence of feedforward projections from rLS to AES (and feedback connections from AES to rLS) (Miceli et al. 1985) and the apparent dominance of AES influences in mediating SC multisensory integration, it seems likely that the final corticotectal output signal of this potential circuit would also come from the AES. There is, however, also the possibility that these effects could be mediated, in part, via corticotectal projections that take an indirect, trans-basal ganglionic route (see Jiang et al. 1997; Miyashita and Tamai 1990; Norita et al. 1986 ). Last, despite the fact that few corticotectal neurons are located on the gyrus of the AES (see Stein et al. 1983), one cannot exclude the possibility that the synergism was due, in part, to the involvement of neurons in this area that would not have been deactivated by cooling AES or rLS alone.
Yet regardless of the specific anatomical circuits or differential contributions necessary for AES and rLS to exert independent or shared control over SC multisensory integration, their designation as “association” cortices seems particularly fitting in the current context. For they proved to be critical for synthesizing, or associating, cross-modal cues. It is interesting to note that, at least for the AES, this role is carried out in two fundamentally different ways: first, intrinsic AES neurons are capable of integrating their multiple sensory inputs in much the same manner as do SC neurons (Jiang et al. 1994a; Wallace et al. 1992); second, as shown here, the corticotectal influences of AES are critical for SC neurons to synthesize cross-modal cues and thereby enhance their multisensory responses. Curiously, however, these tasks are possibly not performed by the same AES neurons: the corticotectal neurons in AES are modality-specific, whereas multisensory AES neurons form a circuit independent of the SC (Wallace et al. 1993). Even the distribution of these two neuronal populations in the AES is distinct. Modality-specific neurons are clustered in the three adjacent areas that form the largely modality-specific representations, SIV, FAES, and AEV. Although a scattering of multisensory neurons can be found in each of these areas (Kimura and Tamai 1992; Minciacchi et al. 1987; Wallace et al. 1992), they are found most frequently at the transitional zones between them (Clemo et al. 1991; Wallace et al. 1992, 1993; but seeJiang et al. 1994b; Korte and Rauschecker 1993). Although somatosensory neurons are already known to be located in rLS (Clemo and Stein 1984; Mori et al. 1991), too little is known about the rLS at present to determine whether its functional organization parallels that of the AES. Presumably, however, the abutment of the multiple sensory cortices (visual, auditory, and somatosensory) that broadly define the location of the rLS also provides a permissive substrate for cross-modal convergence like that present in the AES and in the transition zones of other cortical areas (e.g., see Barth et al. 1995;Ramachandran et al. 1993; Wallace et al. 1992,1993). Given the similar consequences of AES and rLS deactivation on SC multisensory responses, it seems likely that if there are substantial numbers of multisensory neurons in rLS, they, like their AES counterparts, are noncorticotectal. However, these assumptions require empirical validation.
Although the influences of AES and rLS were assessed only in terms of their impact on the response vigor of multisensory SC neurons (seeClemo and Stein 1986; Meredith and Clemo 1989 for discussion of their influences on modality-specific receptive field properties), their functional specificity in this context was particularly striking. Nearly all (i.e., 93%, Fig. 6) of the SC neurons examined depended on one or both of these areas for their ability to engage in multisensory response enhancement, but very few SC neurons depended on these inputs for their ability to respond to modality-specific stimuli. This was evident both at the level of the individual neuron and at the level of the population. The maintenance of the modality-specific responses of single, as well as the population of, multisensory neurons during corticotectal deactivation demonstrates that other tectopetal inputs are fully capable of eliciting reliable modality-specific responses from these neurons. There are many possible candidates for this role. The SC is richly endowed with converging inputs from more than 40 different structures (see Edwards et al. 1979; Huerta and Harting 1984 for reviews). These data, coupled with those from prior physiological (Clemo and Stein 1984; Wallace et al. 1993) and anatomical (e.g., see Stein and Meredith 1993 for a discussion) studies, suggest that many multisensory SC neurons receive parallel sets of converging inputs: on the one hand are those tectopetal projections that descend from neurons in different subregions of association cortex (AES and rLS) and on the other are those projections either ascending from a host of subcortical sources or descending from other regions of cortex. Many different sets of converging afferents can provide the inputs necessary for a neuron to be multisensory, but, thus far only those from AES and rLS have been shown to be capable of rendering these neurons capable of synthesizing their cross-modal inputs.
While the present observations demonstrate that AES and rLS corticotectal influences are critical for SC multisensory integration, the mechanisms underlying their contribution to this process are not immediately apparent. As noted in the preceding text, SC neurons receive convergent inputs from multiple sensory regions at various levels of the neuraxis, and the specific computational contribution made by AES and rLS to the ensemble of convergent signals on any individual neuron is, at present, unknown. Recently, however,Harting and colleagues (1997) have shown that there are substantial morphological differences in the synaptology of the corresponding inputs descending from the somatosensory subdivision of AES and ascending from the trigeminal complex. The possibility has been raised that this distinction in synaptic morphology may distinguish those inputs that are capable of mediating multisensory integration from those that are not. This possibility would be strengthened markedly if it can be shown that these distinctions are also applicable to the ascending and descending components of visual and auditory tectopetal inputs and that the terminal morphologies of AES and rLS corticotectal inputs are distinct from those that characterize other corticotectal systems. The confirmation of this distinction could be quite helpful in developing strategies with which to explore the unique biophysical properties and/or terminal geometries among these different tectopetal projections.
Yet regardless of the specific features that have evolved in the AES and rLS corticotectal projection systems to render them capable of initiating and governing the synthesis of cross-modal signals in SC neurons, the fact that they can do so also gives them dominion over the cross-modal attentive and orientation functions with which multisensory SC neurons are involved. Although the necessity of this corticotectal component adds additional complexity to the scheme underlying SC multisensory integration, it may also provide a substantial benefit. Because many of the functional properties of cortical sensory systems are crafted as a consequence of postnatal experience (e.g., seeBuonomano and Merzenich 1998; Rauschecker 1995,1999) and because cortical neurons are able to impose their experiential adaptations on SC neurons (Wicklegren and Sterling 1969), their inclusion in this circuit may provide a means for modulating the multisensory integrative functions of the SC based on the specific demands of a given environmental circumstance.
We thank Dr. Terrence Stanford for valuable discussions, Dr. Ralph D'Agostino for statistical help, and N. London for technical assistance.
These experiments were supported by National Institute of Neurological Disorders and Stroke Grants NS-22543 and NS-36916.
W. Jiang (E-mail:).
- Copyright © 2001 The American Physiological Society