Orientation Selectivity and the Arrangement of Horizontal Connections in Tree Shrew Striate Cortex William H. Bosking, Ying Zhang, Brett Schofield, and David Fitzpatrick Department of Neurobiology, Duke University Medical Center, Durham, North Carolina 27710 Horizontal connections, formed primarily by the axon collater- als of pyramidal neurons in layer 2/3 of visual cortex, extend for millimeters parallel to the cortical surface and form patchy terminations. Previous studies have provided evidence that the patches formed by horizontal connections exhibit modular specificity, preferentially linking columns of neurons with similar response characteristics, such as preferred orientation. The issue of how these connections are distributed with respect to the topographic map of visual space, however, has not been resolved. Here we combine optical imaging of intrinsic signals with small extracellular injections of biocytin to assess quanti- tatively the specificity of horizontal connections with respect to both the map of orientation preference and the map of visual space in tree shrew V1. Our results indicate that horizontal connections outside a radius of 500 mm from the injection site exhibit not only modular specificity, but also specificity for axis of projection. Labeled axons extend for longer distances, and give off more terminal boutons, along an axis in the map of visual space that corresponds to the preferred orientation of the injection site. Inside of 500 mm, the pattern of connections is much less specific, with boutons found along every axis, con- tacting sites with a wide range of preferred orientations. The system of long-range horizontal connections can be summa- rized as preferentially linking neurons with co-oriented, co- axially aligned receptive fields. These observations suggest specific ways that horizontal circuits contribute to the response properties of layer 2/3 neurons and to mechanisms of visual perception. Key words: orientation selectivity topography optical imag- ing visual cortex surround effects tree shrew biocytin hori- zontal connections Horizontal connections are a prominent feature of the intrinsic circuitry of the visual cortex. These connections originate primar- ily from pyramidal cells, extend for 2���5 mm parallel to the cortical surface, and terminate in a highly selective and patchy manner (Gilbert and Wiesel, 1979, 1983 Rockland and Lund, 1982). A number of experiments have focused on the relationship between the patches formed by horizontal connections and well known modular features of cortical organization such as orientation columns, ocular dominance columns and cytochrome oxidase-rich blobs (Livingstone and Hubel, 1984 T���so et al., 1986 Gilbert and Wiesel, 1989 Malach et al., 1993). The results from these exper- iments suggest that the patchy nature of horizontal connections can be explained by a simple rule: horizontal connections link together select subsets of neurons that share similar receptive field properties. Studies in both cat and monkey visual cortex, for example, have shown that horizontal connections selectively link patches of neurons that have similar orientation preferences (Gil- bert and Wiesel, 1989 Malach et al., 1993). The relationship of horizontal connections to another funda- mental aspect of cortical organization���the orderly map of visual space���is less clear. This issue is of interest because the axon arbors of individual neurons are often elongated across the cor- tical surface, extending further and giving rise to more terminals along one axis of the map than others (Gilbert and Wiesel, 1983, 1989 Matsubara et al., 1985, 1987 McGuire et al., 1991 Kisvarday and Eysel, 1992 Amir et al., 1993, Malach et al., 1993). Furthermore, the results of several physiological and perceptual studies have led to the suggestion that the effects mediated by horizontal connections are not distributed randomly about a point in visual space, but are aligned along an axis that corresponds to a neuron���s preferred orientation. For example, a collinear arrangement of horizontal connections has been implicated in the construction of the elongated receptive fields of layer 6 neurons in cat striate cortex (Bolz and Gilbert, 1989). Likewise, perceptual studies of contour integration and physiological studies of receptive field surround effects in layer 2/3 neurons have provided evidence for facilitatory effects that are much stronger in regions of visual space that lie along the axis of preferred orientation than in regions that lie off this axis (Nelson and Frost, 1985 Fiorani et al., 1992 Field et al., 1993 Polat and Sagi, 1993 Kapadia et al., 1995). Despite the evidence for anisotropic physiological and psychophysical effects, the relationship between the axis of elongation of horizontal connections and the orientation preference of the neurons they interconnect has never been system- atically examined. In the experiments described here, we have combined optical imaging of intrinsic signals with small extracellular injections of biocytin to quantitatively assess both the modular and axial ar- rangement of horizontal connections established by neurons of known orientation preference in layer 2/3 of tree shrew striate cortex. Our results demonstrate a remarkable degree of specificity for both features of horizontal connectivity, and suggest several ways that horizontal connections could contribute to the response properties of layer 2/3 neurons. Received Aug. 5, 1996 revised Dec. 9, 1996 accepted Dec. 12, 1996. This work was supported by National Eye Institute Grant EY06821. We thank Martha Foster for expert assistance with histology and plotting of bouton data, Mike Weliky for assistance with software, and Len White, Michele Pucak, and Justin Crowley for comments on this manuscript. Correspondence should be addressed to William H. Bosking, Box 3209, Depart- ment of Neurobiology, Duke University Medical Center, Durham, NC 27710. Copyright q 1997 Society for Neuroscience 0270-6474/97/172112-16$05.00/0 The Journal of Neuroscience, March 15, 1997, 17(6):2112���2127

MATERIALS AND METHODS Experimental design. Small extracellular injections of biocytin were made in 13 animals. Distributions of labeled boutons resulting from the injec- tions were plotted and analyzed manually (3 cases) or with the assistance of a computer reconstruction system and software routines written in our lab (10 cases, see below for details). Analysis of modular specificity of the bouton distributions was accomplished in 7 of these 10 cases by using optical imaging of intrinsic signals to determine the map of orientation preference in V1 before injection of the biocytin. Analysis of the bouton distributions with respect to the map of visual space was accomplished in all 13 of the cases. Optical imaging was also used to investigate the geometry of the map of visual space in three animals that were not used for analysis of specificity of connections. Animal surgery. Tree shrews were initially anesthetized with a mixture of ketamine hydrochloride (200 mg/kg) and xylazine (4.7 mg/kg) given by intramuscular injection. Atropine sulfate (0.08 mg) was given sub- cutaneously to reduce secretions. An intraperitoneal cannula was in- serted, the trachea intubated, and the animal was placed in a modified stereotaxic frame allowing unobstructed viewing of the stimulus monitor. During surgery, anesthesia was maintained with a 2:1 mixture of N2O/O2 supplemented with 2% halothane. Body temperature was maintained with a thermostatically controlled heating blanket. The eyes were kept moist by using planar contact lenses. An incision was made in the scalp, muscle and fascia reflected, and the bone overlying visual cortex was thinned by scraping with a fine scalpel. Wound margins and incisions were treated with a long-acting local anesthetic (bupivacaine) and pressure points were treated with a lidocaine ointment. The animal was paralyzed using pancuronium bromide (0.8 mg initial dose for first 1.5 hr, then 0.2 mg/hr) administered through the intraperitoneal cannula to prevent eye movements, and artificially respired at a rate and volume sufficient to maintain expired CO2 at 3���4%. During optical imaging, halothane levels were reduced to 1%, and the N2O/O2 mixture to 1:1. Any signs of distress evident in the electrocardiogram or expired CO2 were treated by imme- diately increasing the level of halothane. Optical imaging. Optical imaging of intrinsic signals was accomplished using an enhanced video acquisition system (Optical Imaging Inc.) ap- plying techniques similar to those of Grinvald and colleagues (Grinvald et al., 1986 Bonhoeffer and Grinvald, 1991, 1993). Images were obtained directly through the thinned bone overlying the V1 area. The cortex was illuminated with orange light (605 nm) and visualized with a tandem lens macroscope attached to a low noise video camera. Visual stimulation for optical imaging was provided by a separate stimulus computer (386 PC with SGT1 graphics board and STIM software provided by Kaare Chris- tian). The stimuli used consisted of high-contrast square wave gratings (6.258 dark phase, 1.258 light phase, drifted at 22.58/sec). Four gratings of orientation 08, 458, 908, and 1358 with respect to horizontal were used. Each grating was moved back and forth along an axis that was orthogonal to the orientation of the grating. Data were acquired during between 20 and 80 presentations of each stimulus. The summed images acquired during the presentation of one grating were subtracted from the summed images acquired during presentation of the orthogonal grating to create differential maps of orientation preference (difference images), (Blasdel, 1992). Difference images were 655 3 480 pixels in resolution, with either 62 or 75 pixels per millimeter depending on the lens combination used. Resulting difference images were smoothed using a 7 3 7 pixel mean filter kernel. Low frequency noise was reduced by convolving the image with a 40 3 40 pixel mean filter kernel and subtracting the result from the original image. Difference images were normalized by dividing the devi- ation from the mean at each pixel by the average absolute deviation across the entire image (Weliky et al., 1995). Finally, vector summation of the difference images was done on a pixel by pixel basis to create a color coded orientation preference map (Bonhoeffer and Grinvald, 1991,1993 Blasdel, 1992). Optical imaging of the geometry of the map of visual space in V1 was accomplished using procedures similar to those developed by Campbell and Blasdel (1995). The technique uses difference imaging for spatial location to identify areas of cortex that respond preferentially to stimu- lation of a particular line in visual space. The stimuli used consisted of two gratings of the same orientation, each with a period of 108 but differing in the width and spatial location of the light phase of the grating (grating 1 5 18 light phase, 98 dark phase grating 2 5 38 light phase, 78 dark phase). The two gratings were placed and moved in such a way that the light phases of the gratings covered non-overlapping regions of the stimulus monitor (grating 1 moved 618 from original position, grating 2 moved 628 from original position). Images acquired during presentation of grating 2 were subtracted from images acquired during presentation of grating 1 for each orientation to create topographic difference images. Topographic difference images were smoothed with a 7 3 7 pixel mean filter kernel for presentation. At the conclusion of the optical imaging phase of the experiment, a reference image of the surface blood vessels was acquired, while the cortex was illuminated with green (540 nm) light. Biocytin injections. Injection sites were selected by examining the dif- ference images and orientation preference maps to find an area of relatively constant orientation preference. Selected sites were located using blood vessels visible in the reference image and on the cortical surface as seen under a surgical microscope. The orientation tuning of the injection site was confirmed by recording multi-unit activity through the injection micropipette. Tuning curves were obtained by averaging re- sponses from 5���8 trials of 9 oriented stimuli. Iontophoretic injections were made with glass micro pipettes with a tip diameter of 10 mm containing 5% biocytin (Sigma, St. Louis, MO) in saline using pulsed current (7 sec on, 7 sec off) of 2.5 mamps for 10���15 min. After the injection the animal was sutured, recovered from paralytic and anesthe- sia, and returned to its cage. Tissue processing. After a 16 hr recovery period, the animal was deeply anesthetized with Nembutal (25 mg, i.p.) and transcardially perfused with 0.9% saline, followed by 10% formalin in 0.1 M sodium phosphate buffer. The brain was removed and a block of cortex containing V1 was flattened while immersed in 20% sucrose in 0.1 M sodium phosphate buffer and maintained at 48C overnight. The following day 40 mm tangential sections were cut from the block on a freezing microtome. Care was taken to collect a first section that contained outlines of the surface vasculature, so that the tissue sections could later be aligned with the reference image. Our procedures for visualization of the biocytin label have been published (Usrey and Fitzpatrick, 1996). Briefly, goat anti-biotin and biotinylated rabbit anti-goat antibodies (Vector Laboratories, Burlingame, CA) were used to amplify the signal before processing with the standard avidin��� biotin complex (Vectastain Kit PK-4000, Vector, Burlingame, CA) reac- tion, and diaminobenzidine with nickel and cobalt intensification. Visu- alization of the V1/V2 border was accomplished by using a Nissl counterstain. Some sections were counterstained before they were cov- erslipped, and some sections were counterstained after bouton distribu- tions were plotted. Bouton plotting, alignment, and analysis. Blood vessel outlines in the first tissue section, radial vessel profiles, and labeled boutons from 2���6 sections were plotted using Neurolucida software (Microbrightfield, Colchester, VT). In each case, enough sections were plotted to assess the bouton distribution in the superficial layers throughout the dorsal region of V1. The data were stored as a series of x,y coordinates for each bouton or each blood vessel reference point in the slice. These data were aligned with the optical imaging data using software routines written as an extension to the public domain National Institutes of Health Image program [original version written by Wayne Rasband at National Insti- tutes of Health and available from Internet by anonymous ftp from zippy.nimh.nih.gov or on floppy disk from NTIS (5285 Port Royal Rd., Springfield, VA 22161, part number PB93���504868)]. The first stage of the alignment was to align the computer drawing of the first tissue section to the reference image acquired during optical imaging. Global scaling, rotation, and x,y translations were applied to the stored coordinates to achieve the best overall alignment possible, with emphasis placed on the area of the slice containing the injection and bouton data. The second stage of the alignment was to align deeper tissue sections, containing labeled boutons, to an overlay of the first section drawing and reference image by using the profiles of blood vessels that course radially through V1. As in the first stage, only global scaling, rotation, and translations were used to align the section. At the end of this procedure, the same transformations used to align the blood vessel reference points in deep sections were then applied to the bouton data (coordinates). This allowed direct comparison of the bouton distributions with the map of orientation preference, which was in the exact same field of view as the reference image. Bouton distributions were not compared with the map of orien- tation preference until the completion of the alignment process. Bouton tuning curves were obtained by counting the number of boutons that contacted sites with various orientation preferences using 108 bins of orientation preference. Bouton tuning curves were computed separately for boutons that were greater or less than 500 mm from the injection site. Axial specificity was assessed by counting the number of boutons within each 108 sector around the injection site, excluding the boutons that were within 500 mm of the injection site. Profiles showing the number of Bosking et al. ��� Specificity of Horizontal Connections in Striate Cortex J. Neurosci., March 15, 1997, 17(6):2112���2127 2113

boutons versus distance were obtained for both preferred and orthogonal axes (6308). Maximum distance of projection along the preferred and orthogonal axes were determined by calculating the maximum distance that a minimum density of 40 boutons/0.01 mm2 could be found. Three of the cases used for determination of axial specificity were plotted using an alternative system that did not allow the use of the extended National Institutes of Health Image routines for alignment and analysis. For these cases, the bouton distribution was plotted from super- ficial sections containing layer 2/3, and the V1/V2 border was plotted in thionin stained sections from layer 4. The border was then transferred to the layer 2/3 plots by using the position of radial blood vessel profiles. The number of boutons within 108 sectors around the injection site was then counted manually. Similar results were obtained by both methods, but density, distance profiles, and maximum distance calculations were not attempted using the manual counting method. RESULTS In the next sections, we describe the results from experiments designed to examine the distribution of biocytin-labeled terminals with respect to the map of orientation preference and the map of visual space in striate cortex (V1) of the tree shrew. In the first section we begin by considering the basic features of the orienta- tion preference map that are relevant for understanding the quantitative assessment of modular specificity. Maps of orientation preference in tree shrew striate cortex Previous studies demonstrated that neurons in the superficial layers of tree shrew striate cortex are sharply tuned to the orien- tation of edges (Humphrey et al., 1980a) moreover, on the basis of 2-deoxyglucose (2-DG) labeling, it was suggested that neurons with similar orientation preferences were arranged in a series of parallel bands or stripes that intersected the V1/V2 border at right angles (Humphrey et al., 1980b). If the 2-DG experiments have provided an accurate picture of the layout of iso-orientation domains, then the map of orientation preference in the tree shrew would be considerably different from what has been described in monkeys and cats (Bonhoeffer and Grinvald, 1991, 1993 Blasdel, 1992). However, the difference could also be attributed to the fact that the results in these other species are based on more recently developed optical imaging techniques. These techniques provide a more detailed assessment of the map because they permit the comparison of the patterns of activity evoked by multiple stimulus orientations in the same region of cortex. A complete analysis of the map of orientation preference in the tree shrew striate cortex based on optical imaging techniques will be presented elsewhere here we simply describe the basic features of the map and em- phasize that the arrangement of orientation preference maps in the striate cortex of the tree shrew, monkey, and cat are funda- mentally similar. Orientation selectivity was also observed in V2 of some animals, although the signal strength in V2 was much weaker for reasons that remain unknown. The functional organi- zation of orientation selectivity and connections in V2 are not explored in this paper. Our optical imaging experiments confirm some of the features described in the 2-DG experiments. For example, in almost all of the orientation difference images, we found regions of the cortex that had the appearance of parallel alternating dark and light stripes (Fig. 1A). These stripes were common in the most caudal part of the exposed surface and along the V1/V2 border where they intersected the border at right angles. However, other parts of the map were decidedly less stripe-like in appearance in the center of the exposed surface, for example, the stripes were often replaced by a less regular and more punctate set of domains. Furthermore, a comparison of the patterns of activity evoked by different stimulus orientations revealed that even those regions of the map that had a stripe-like appearance in images generated for one stimulus orientation would often appear to break up into discontinuous patches in images generated for other stimulus orientations. Taken together, these observations suggested that the map of orientation preference in the tree shrew is far more complex than was indicated by the earlier 2-DG experiments. The fine-scale mapping of orientation preference is best appre- ciated by combining the individual difference images using vector summation to create an orientation preference map where colors are used to represent the preferred orientation at each site (Fig. 1B). This analysis confirms that the map of orientation preference in the tree shrew has the same basic organizational features that have been described previously in monkey and cat striate cortex (Bonhoeffer and Grinvald, 1991, 1993 Blasdel, 1992). In many regions of the map, commonly referred to as pinwheels, a contin- uous shift in orientation preference is obtained by sampling around a point, or singularity. Examples of two pinwheels from the orientation preference map shown in Figure 1B are shown at higher magnification in Figure 1C. Linear zones, regions of the map in which a progressive change in orientation preference is obtained by sampling along a straight line (Blasdel, 1992 Ober- mayer and Blasdel, 1993), are also a prominent feature of orien- tation preference maps in the tree shrew. As predicted from the difference images, linear zones are common along the V1/V2 border and along the caudal edge of the dorsal portion of V1. They can extend for 2���3 mm, a distance that covers several full repeats of the orientation cycle. In Figure 1, A and B, an especially large linear zone is visible in the caudal portion of the map, and this same linear zone has been enlarged in the left hand side of Figure 1C. Modular specificity of horizontal connections To assess the modular specificity of horizontal connections, com- bined optical imaging and biocytin injection experiments were accomplished in seven animals. After the optical imaging phase of the experiment, an injection site was selected and the orientation tuning of the injection site was confirmed by recording multi-unit activity through the injection pipette. Our biocytin injections resulted in the labeling of a small number of cell bodies (12���65) that were confined to sites that were typically less than 200 mm in diameter (average diameter 176 mm, largest diameter 320 mm see Fig. 2A,B). In two of our cases, one or two retrogradely labeled cells could be found at some distance from the injection site, but this did not hamper our ability to detect the underlying specificity of the connections. The axonal processes of the labeled neurons were well labeled (Fig. 2C) and exhibited the characteristic bou- ton terminal swellings that have been described in other species (Gilbert and Wiesel, 1983 Amir et al., 1993 Kisvarday and Eysel, 1992). As seen in Figure 2A, labeled axons extended away from the injection site for several millimeters and gave rise to promi- nent patches. Borders of bouton patches were determined subjec- tively and by thresholding a density plot of the bouton distribu- tions. The two methods were in good agreement and the average size of bouton patches was determined to be 400 mm 3 250 mm. This information is provided for comparison with other reports only the borders of bouton patches were not used in the analysis of modular or axial specificity. The distribution of labeled boutons was plotted from between two and six sections for each case. The interpretation of our findings rests on the accuracy with which we are able to align plots of labeled boutons from the anatomical sections with optical maps of orientation preference. 2114 J. Neurosci., March 15, 1997, 17(6):2112���2127 Bosking et al. ��� Specificity of Horizontal Connections in Striate Cortex