Functional-Specic Projections from V2 to V4 in Macaque Monkeys

Previous studies have revealed modular projections from area V2 to area V4 in macaques. Specically, V2 neurons in cytochrome oxidase (CO)-rich thin and CO-sparse pale stripes project to distinct regions in V4. However, how these modular projections relate to the functional subcompartments of V4 remains unclear. In this study, we injected retrograde uorescent tracers into different functional domains (color, orientation, and direction) in V4 that were identied with intrinsic signal optical imaging. We examined the labeled neurons in area V2 and their locations with respect to the CO patterns. Covariation was observed between the functional properties of the V4 injection sites and the numbers of labeled neurons in particular CO stripes. This covariation indicates that the color domains in V4 mainly received inputs from V2 thin stripes, whereas V4 orientation domains received inputs from pale (major) and thick (minor) stripes. Although there are motion-sensitive domains in both V2 and V4, our results did not reveal a functional-specic feedforward projection between them. These results conrmed previous ndings of modular projections from V2 to V4 and provided functional specicity for these anatomical projections. Together, these ndings indicate that color and form remain separate in the ventral midlevel visual processing.


Introduction
Area V2 in primates contains functionally distinct subcompartments that can be revealed with cytochrome oxidase (CO) staining. Thin CO stripes contain more color-sensitive neurons, thick and pale stripes contain more orientation neurons, and thick stripes contain more binocular disparity and direction neurons (DeYoe and Van Essen 1985;Livingstone 1985, 1987;Peterhans and von der Heydt 1993;Levitt et al. 1994;Shipp and Zeki 2002). Functional imaging also con rmed such functional segregation in V2 (Ts'o et al. 1990(Ts'o et al. , 2001Roe and Ts'o 1995). These CO stripes also exhibit different patterns in projections to downstream areas. Neurons in thin and pale stripes mainly project to area V4, whereas neurons in thick stripes mainly project to area MT (DeYoe and Van Essen 1985;Shipp and Zeki 1985;Nascimento-Silva et al. 2003). In V2-V4 feedforward projections, thin and pale stripes appear to project to separate regions in V4 (Zeki and Shipp 1989;Nakamura et al. 1993;DeYoe et al. 1994;Felleman et al. 1997;Xiao et al. 1999), indicating a modular organization of V4 and modular connections with V2 functional organizations. However, the functional properties of these V2-receiving modules remain unclear. With the discovery of different types of functional modules in V4 (Ghose and Ts'o 1997;Conway et al. 2007;Tanigawa et al. 2010), it is important to establish the relationship between the anatomical and functional modules in V4.
Area V4 is an important visual area that performs midlevel visual processing (Roe et al. 2012;Pasupathy et al. 2020). Similar to its upstream areas, V4 also contains functional modules that preferentially process color and form information (Ghose and Ts'o 1997;Conway et al. 2007;Tanigawa et al. 2010).
These spatially segregated modules may receive inputs from the corresponding color and form modules in V2. One of the purposes of this study was to provide direct evidence to support these functional connections between V2 and V4. Another potential functional connection we wanted to investigate was motion-sensitive modules between these two areas. Previous studies have shown that in both V2 ) and V4 (Li et al. 2013), direction-selective neurons tend to cluster and form functional modules.
These ndings inspired us to investigate the functional link between these two modules.
In this study, we rst mapped 3 functional responses in V4 with intrinsic signal optical imaging. We then injected retrograde tracers into different V4 regions with different functional properties. Finally, we counted labeled neurons in different CO compartments in V2. We observed covariations between the labeled neurons in V2 and the functional properties of the tracer injection sites in V4; we then used this information to estimate the connections between V2 and V4 functional compartments.

Materials And Methods
Animal subjects A total of 3 adult male macaque monkeys (2 Macaca mulatta, 1 Macaca fascicularis) were examined. All procedures were performed in accordance with the National Institutes of Health Guidelines and were approved by the Institutional Animal Care and Use Committee of Beijing Normal University.

Surgery
Aseptic surgery procedures were similar to those described in previous papers (Li et al. 2013). Animals were arti cially ventilated and anesthetized with iso urane (1.5-3%) in oxygen. Heart rate, end-tidal CO 2 , blood oximetry, and body temperature were closely monitored, and anesthesia depth was estimated based on these values. A circular craniotomy (24 mm in diameter) was performed over area V4. The center of the craniotomy was 30-38 mm from the midline and 10-18 mm from the posterior bone ridge. A durotomy was performed to expose parts of areas V1, V2 and V4 (illustrated in Figure 1a&b).

Intrinsic signal optical imaging
Intrinsic signal optical imaging (ISOI) was performed immediately after surgery. The imaging procedures used were described in detail in previous papers (Li et al. 2013). The brain was stabilized with a cover glass (2-3 mm thickness) and agar (4%). The anesthetic was switched to propofol (induction 5 mg/kg, maintenance 5 mg/kg/hr, i.v.). To prevent eye movements, animals were paralyzed with vecuronium bromide (induction 0.25 mg/kg, maintenance 0.05 mg/kg/hr, i.v.) and respirated. Atropine sulfate (10 mg/ml) or tropicamide (5 mg/ml) was applied to dilate the pupils. Appropriate contact lenses were used to focus the eyes on a CRT screen 57 cm from the eyes. The cortex was illuminated with 632 nm light, and the re ectance light was collected by Imager 3001 (Optical Imaging Inc, Germantown, NY) at a 4 Hz frame rate. The frame size was 654´540 pixels, with a pixel size of 38 microns. Each imaging session started 0.5 seconds before the visual stimulus and lasted for 4 seconds (16 frames). The duration of the visual stimulation was 3.5 seconds. Intertrial interval was 8 seconds.

Visual stimulus
The visual stimuli used in this study were the same as those previously used to obtain functional maps (Lu et al. 2010 for V2 maps;Li et al. 2013 for V4 maps). Visual stimuli were created using ViSaGe (Cambridge Research Systems Ltd) and displayed on a 21-inch CRT monitor refreshing at 100 Hz. The stimulus patch was a square larger than 10 degrees, presented binocularly, and that fully covered the visual eld of the imaged cortices. To obtain 3 types of functional maps (color, orientation, and direction), two sets of gratings were used. For the color functional map, responses to red-green isoluminant sinewave gratings and black-white sinewave gratings were compared. The spatial frequency (SF) of the gratings was 1 cycle/degree, and the temporal frequency (TF) of the gratings was 4 cycle/second.
Gratings were presented in one of two orientations (45° or 135°) and in random directions orthogonal to the grating orientation. Data from different orientations were pooled. For orientation and direction maps, black-white square wave gratings were used. Gratings had an SF of 1 cycle/degree, had a TF of 4 cycles/second and drifted in one of 8 equally spaced directions (0°, 45°, 90°, 135°, 180°, 225°, 270°, or 315°). All stimuli were presented in a random order. An orientation map was obtained by comparing responses to orthogonal gratings (direction pooled), and a direction map was obtained by comparing responses to oppositely moving gratings.
Functional map guided tracer injection V4 injection sites were selected from the 3 functional maps (color, orientation, and direction) with the aim of maximizing the response differences. Three types of tracers (CTB-488, CRB-555, and CTB-647) (1% Invitrogen) were used in this study. A pressure injection system (53311, Stoelting) and a glass pipette with a tip size of 23-35 mm (beveled) were used. The connection tube was lled with silicon oil, and the tracer was sucked into the glass pipette from its tip. A small amount of air (~40 nl) was sucked into the tip. The pipette was lowered to the planned V4 site based on the surface blood vessel pattern. The pipette tip was lowered into the cortex to a depth of 0.8-1.2 mm, and then was withdrawn 0.1 mm; next, the tracer injection was started. The amount of injection (30-100 nl, see Table 1) was measured starting at the air disappearing from the visible tip. After injection, the pipette was held in place for 10 min before retraction. The withdrawal distance of the pipette was recorded as the depth of the injection (see Table 1) when the tip left the cortex. One to 3 injections were given to each animal (different tracers for different functional sites). An imaging chamber (Li et al. 2013;Tang et al. 2020) was implanted. In vivo uorescent images were examined (not shown) one week after the tracer injections to con rm tracer loading and to determine the center of the tracer diffusion.

Histology and areal delineation
Eighteen to 23 days after the tracer injections, the animals were sacri ced with an overdose of sodium pentobarbital (80-100 mg/kg) and perfused transcardially (step 1: 2 L of phosphate buffer, step 2: 1 L buffer with 2% paraformaldehyde, and step 3: 1 L buffer containing 2% paraformaldehyde and 10% sucrose; all pH 7.4). The brain tissue containing the lunate sulcus was separated, and the lunate was attened. The tissue was post xed in buffer containing 2% paraformaldehyde and 30% sucrose at room temperature for 2-3 h. Then, the tissue was cryoprotected in 30% sucrose buffer overnight at a temperature of 4°C. The attened tissue was sectioned horizontally on a freezing microtome Yamato). Several hole markers were made with a used microelectrode perpendicular to the surface (with 5 mm intervals) to aid subsequent alignments. The rst slice, which contained surface blood vessels, was sectioned at a thickness of 120-150 mm, and the rest of the slices were 60-80 mm. Most of the slices were loaded onto a glass slide and sealed with cover glass for uorescent imaging. For monkeys A, B and C, 3 (out of 11), 3 (out of 14), and 1 (out of 10) slices, respectively, were selected for CO staining (Wong-Riley 1979) and imaging (MVX10, Olympus, 1X objective).
For uorescent imaging, we used a confocal microscope (A1R, Nikon) and a 10X objective. To determine the structure of labeled neurons, the software parameters (laser power, PMT ampli cation, offsets) were adjusted to make the labeled neuron su ciently bright but not overexposed. For CTB-488, CTB-555, and CTB-647, we used 500-550 nm, 570-620 nm and 663-738 nm lters, respectively. These monochromatic images were colored green, red, and blue in software (Nis-Elements). These 3 images were then merged.
In the merged images, neurons were identi ed manually based on their color: red, green, or blue neurons indicated CTB-488, CTB-555, or CTB-647, respectively; yellow, cyan, or magenta neurons indicated doublelabeled neurons. We did not include triple-labeled neurons (white) since some non-neuronal noise (blood vessels and dust) also appeared white. These triple-labeled neurons accounted for less than 1% of the total labeled neurons. In the confocal images, the overexposure area around the injection site (due to the high density of labeled neurons) was determined as the size of the injection area (Table 1). Fluorescent label and CO staining images from adjacent slices were aligned according to the hole markers we made (imtransform, MATLAB). The ISOI maps and images of the rst slice were aligned based on the surface blood vessel pattern. To determine the CO stripe, typically all the CO staining images of the same monkey were merged (monkeys A and B). CO stripes were identi ed based on the width of the densely stained stripes and their alternative patterns in a CO cycle. These slices were no longer suitable for uorescent imaging. To obtain an accurate cell count, the labeled neuron numbers of these slices were estimated based on the averages of labeled neurons in their neighboring slices.

Data analysis
Functional maps V4 functional maps (e.g., Figure 1d-f) were obtained with established SVM methods (Xiao et al. 2008;Chen et al. 2016). Compared with regular subtraction maps, SVM maps have a better signal/noise ratio while maintaining the linear relationship of the map signals. For each stimulus condition, we rst obtained a dR/R image using the following formula: dR/R = (R 6-14 -R 1-3 )/R 1-3 , in which R 6-14 is the average of frames 6-14 and R 1-3 is the average of frames 1-3. Two sets of dR/R images (e.g., from color and luminance stimuli) were then used to train an SVM classi er (Libsvm, Chang and Lin 2011). An optimal classi er was obtained after 5-fold cross-validation. The weight map of the trained SVM classi er was used as the functional map (e.g., Figure 1d-f), in which a pixel value represents the contribution of the pixel to the classi cation. Each map was bandpass ltered (high cutoff: 0.1 mm/cycle, low cutoff: 3 mm/cycle). using the maximum value of each pixel. A direction strength map was obtained from the 4 direction maps with the same method. Each of the 3 functional strength maps was then ltered with a 1.5 mmdiameter circular mean lter and normalized to 0-1 (Figure 1g-i). The ltering procedure mimicked tracer diffusion (average diameter 1.5 mm). The pixel values of the 3 functional strength maps were then used as the functional strength values of the V4 injection sites in the following data analysis. In one case (monkey C), we also obtained 2 functional strength maps (orientation and direction) from area V2 and analyzed the spatial relationship between V2-labeled neurons and these strength maps (Figure 5g).

Correlation Analysis and linear regression
Spearman correlation (corr, MATLAB) was calculated between V4 functional strength values and numbers of V2-labeled neurons (Figure 3a-i), as well as between V4 functional strength values (Figure 4eg). A linear regression (poly t, MATLAB) was also calculated for each of these pairs. The goodness of t was evaluated with R 2 .

Results
A total of 3 hemispheres of 3 macaques were used in this study. For each hemisphere, color, orientation and motion direction maps were rst obtained in anesthetized conditions with intrinsic optical imaging (Li et al. 2013). V4 regions with different functional properties were selected and used for tracer injection. For monkeys A and B, 3 injections of different tracers (CTB-488, 555, 647) were administered in one hemisphere for each animal. For monkey C, only one injection (CTB-647) was administered. Two to 3 weeks after injection, the animals were sacri ced. Area V2 was attened, sliced and examined for labeled neurons under a confocal microscope. Some of these slices were stained and examined for CO compartments under a light microscope.
Functional properties of the V4 injection sites was located in a color domain, which also had certain directional preferences. Each strength map ( Figure   1g-i) was ltered with a circular mean lter whose diameter equaled the injection area, and the pixel value of the injection site (single pixel) was used as the functional strength for that injection site. Thus, for each injection, we obtained 3 functional strength values: color, orientation, and direction. The functional strength distributions were different for the 3 injections in monkey A (Figure 1j). Next, we examined whether such differences cause systematic changes in labeled neurons in V2.

Retrograde labeled neurons in V2
Eighteen to 23 days after the V4 injection, animals were sacri ced. Area V2 was attened and cut into slices 60-80 mm thick. For each case, 1-3 slices were selected and used for CO staining to determine the CO stripe pattern in V2 (Figure 2b). Confocal uorescent images were obtained from the rest of the slices (e.g., Figure 2c&d). Labeled neurons were manually identi ed from each slice and marked on a stripeidenti ed image (Figure 2e) after slice alignment. For each color (tracer), the labeled neurons formed several bands perpendicular to the V1/V2 border. In this case, CTB-488 (green)-and CTB-647 (blue)labeled neurons were mostly located in thin stripes, whereas CTB-555 (red)-labeled neurons were mostly located in pale stripes (Figure 2e). Some labeled neurons were seen in thick stripes (mostly on their edges), but all 3 tracers were largely absent from thick stripes. In addition, we also observed that more super cial-and deep-layer neurons were labeled than those in the midlayers (not shown). For each tracer, we calculated the distribution of its labeled neurons among 3 stripe types (Figure 2f). To compare distributions among different tracers and cases, we used percentages of neurons instead of neuron numbers. CTB-647 (blue) was injected into a V4 region close to a color domain (Figure 1g), and its labeled neurons in V2 were mostly found in thin stripes (Figure 2f left). CTB-555 (red) was close to a V4 orientation domain (Figure 1h), and its V2 labeled neurons were mostly found in pale stripes (Figure 2f middle). For CTB-488 (green), the V4 injection site was close to a direction domain (Figure 1i), and its labeled neurons in V2 were mostly found in thin stripes (Figure 2f right). Thus, there appear to be certain correlations between the V4 injection site (and its functional properties) and the distribution of labeled neurons in V2. This qualitative observation is examined quantitively below. In total, we performed 7 injections in 3 hemispheres of 3 monkeys. Detailed information on the injection sites and labeled neurons in V2 is shown in Table 1.

V2-V4 correlation analysis
To analyze the relationship between the functional strengths of the V4 injections and the distributions of labeled neurons in V2 CO stripes, scatter plots of these two factors were used (Figure 3 a-i). Each plot comprises 7 data points, representing 7 tracer injections. For each injection, the X axis value represents its functional strength in one functional map, and the Y axis value represents its percentage of labeled V2 neurons in a certain stripe type. The dotted line is a linear regression of these data points. Figure 3a shows that the percentages of labeled neurons in thin stripes were positively correlated with the color strength of the V4 injection sites. That is, when the injection site was closer to the center of the V4 color domains, more labeled neurons were found in the V2 thin stripes. Such a positive correlation was not present in V2 pale ~ V4 color (Figure 3b) or V2 thick ~ V4 color (Figure 3c) plots. Taken together, these results indicated that V4 color modules tended to receive inputs from V2 thin stripes.
The V4 orientation feature had a positive correlation with V2 pale and thick stripes (Figure 3 e&f). The slope was steeper for pale stripes than for thick stripes, indicating a stronger contribution to V4 orientation domains from pale stripes than from thick stripes. Figure 3g shows that the V4 direction had a positive correlation with V2 thin stripes. This is probably because many direction domains we selected for injection also had strong color responses (see Discussion). Figure 3h&i shows that the V4 direction had a negative correlation with both pale and thick stripes, indicating that the V4 direction domains did not tend to receive input from V2 pale or thick stripes. Since V2 direction neurons were more likely to be found in thick stripes, V4 direction domains are not likely to be a main efference for V2 motion. Further evidence from V2 imaging also supports this conclusion ( Figure 5). For each plot, a Spearman correlation coe cient was calculated and illustrated in Figure 3j.
The slopes of the 9 plots in Figure 3a-i are not independent from each other. One reason is that the functional maps in V4 have certain spatial correlations. For example, the color and orientation domains in V4 tend to occupy complementary regions Tanigawa et al. 2010). We analyzed functional maps in 12 V4 cases, including 3 cases in this study and 9 additional cases. The results con rmed previous ndings and showed a negative relationship (Figure 4a). In addition, the selection of tracer injection sites is also important. In this case, the color and orientation properties of the 7 injection sites also showed a negative correlation (Figure 4e), consistent with the map features. In Figure 3, the slopes in the rst and second rows are opposite to each other, which may re ect this negative relationship. In contrast, the V4 direction was positively correlated with V4 color (Figure 4b) and with V4 orientation (Figure 4c), consistent with previous ndings (Li et al. 2013). Another factor that affects the independence of the scatter plot slopes is that V2 neuron numbers were normalized within a tracer group (to percentages). Thus, for 3 plots in a row, their slopes would not be the same sign (i.e., all positive or all negative). Normally, a weak positive slope or a zero slope may become negative because of this factor.
Thus, a positive correlation (r > 0) indicates that two factors were positively correlated. If two factors had no correlation (r = 0), the slope should be at instead of negative. A negative correlation (r < 0) usually was due to some other factors (e.g., the negative correlation between V4 functional maps). Therefore, a positive correlation in Figure 3 was explained as the existence of a corresponding V2-V4 connection. A negative correlation was normally explained as an absence of a corresponding V2-V4 connection after taking into account multiple factors, including the spatial relationships between functional maps in V4 (Figure 4a-d), the properties of V4 injection sites (Figure 4e-h), and the normalization procedure for V2 neuron numbers.
In macaque monkeys, area V2 is mostly buried in the lunate sulcus, which limits the applicability of optical imaging. Thus, in this study, we mainly relied on CO staining to determine the V2 functional compartment. In monkey C, we also obtained functional maps of the V2 region containing labeled neurons (Figure 5g) so that we could determine the functional properties of these V4-projecting neurons in addition to their CO properties. We only obtained V2 orientation and direction maps. This case only had one injection in V4, which had dual response features (color and direction) (Figure 4b&c). However, the labeled V2 neurons were mostly found in regions having weak direction and weak orientation responses (Figure 4 g, h, and i). Thus, this evidence con rmed earlier ndings based on V2 CO stripes that V4 color and direction domains did not tend to receive inputs from V2 thick or pale stripes. This further suggests that the V4 direction domains do not receive direct input from the V2 direction domains. Figure 6 summarizes our ndings. In V2-V4 projection, V2 thin stripes mainly project to V4 color modules; V2 pale stripes mainly project to V4 orientation modules, and V2 thick stripes have much fewer projections to V4 than the other two stripes and mainly project to orientation modules. Pale and thick stripes do not project to V4 direction modules. It is not clear whether thin stripes project to V4 direction modules.

Discussion
In this study, we injected retrograde tracers into different functional modules in V4 and examined the patterns of labeled neurons in V2. We observed functional-speci c projections from V2 CO stripes to V4 functional modules. In this study, functional and projection factors were quanti ed as was the correlation between these two factors. The results not only con rmed previous anatomical ndings but also established a relationship between V2 CO stripes and V4 functional modules.

Parallel visual processing
The existence of functional modules, as well as the parallel projections among these modules, indicate that the process of visual information occurs in parallel. In V2-V4 projections, previous work has found that thin and pale stripes project to different regions in V4, whereas thick stripes barely project to V4 (DeYoe and Van Essen 1985;Zeki and Shipp 1989). The same projection patterns were also observed in New World monkeys (Nascimento-Silva et al. 2003, 2014. Our ndings are generally in line with these earlier ndings. The additional information we provide is the functional properties of these V4 targets. For orientation projection from V2 to V4, our data show that in addition to pale stripes, some thick neurons (mostly those located on the edges of thick stripes) also projected to V4 orientation domains. Our ndings generally support the view that visual features are processed in parallel in the visual pathway.
However, much evidence has also shown that different visual features (color, form, motion, and depth) are not processed in parallel. A large degree of integration has been observed at both the single-cell and functional architecture levels. For example, many V2 neurons exhibit dual selectivity for color and orientation (Gegenfurtner et al. 1996;Shipp et al. 2009). Orientation and direction neurons are also found in thin stripes, and color neurons are found in pale and thick stripes (Gegenfurtner 2003). In general, it appears that parallel processing is more evident at the functional architecture level than at the single-cell level. This means that the same types of information (e.g., orientation) processed in different functional modules (e.g., pale and thick stripes) is eventually sent to different downstream areas (e.g., V4 or MT). The functional signi cance of this processing strategy is worth further investigation.

Origin of V4 direction selectivity
Thus far, we have a relatively clear idea about the information sources for V4 color and orientation features. It is still unclear which inputs to V4 carry direction-selective motion information. It has been shown that many (~15%) V4 neurons have direction selectivity, some of which cluster into direction domains (Li et al. 2013). One potential region that provides input is V2, where direction neurons are mostly located in thick stripes. However, our results show that V4 direction domains do not receive inputs from V2 thick or pale stripes (Figure 3 h&i). In addition, our preliminary imaging results also show that V4 direction domains do not receive inputs from V2 direction domains (Figure 5 g&h). Thus, it appears unlikely that V4 direction domains obtained their motion inputs from V2 thick or pale stripes. For the positive correlation between the V4 direction and V2 thin stripes (Figure 3g), one possibility is that a major input to the V4 direction domain is V2 thin stripes, and V4 direction neurons indeed receive direction inputs from V2 thin projections. This possibility sounds less likely but not unreasonable, since there are direction neurons in V2 thin stripes, although they are fewer than those in thick stripes (Peterhans and von der Heydt 1993;Gegenfurtner et al. 1996). Another more likely possibility is that the phenomenon that V4 direction domains receive inputs from V2 thin stripes is an artifact. This is because the V4 injection sites we selected mostly had positively correlated color and direction strengths ( Figure  4f), and it was V4 color neurons (not direction neurons) that received inputs from V2 thin stripe projections. In other words, we needed more differences in color and direction selectivity in our injection sites to differentiate their inputs. Compared with the V4 color domains, the clustering of the V4 direction domains was weaker (fewer neurons and less degree of clustering), and the tracing results were more likely dominated by the connection patterns of color neurons. If this is true (V4 direction domains do not receive inputs from V2 thin stripes), then V2 is unlikely the source of V4 direction selectivity.
Another potential source for V4 direction selectivity is area MT, since there are direct projections from MT to V4. However, a previous study selectively inhibited M or P layers in the LGN and found that V4 direction selectivity does not particularly rely on M or P input. This is different from what was observed in MT neurons, whose response mainly relied on M inputs (Maunsell et al. 1990;Ferrera et al. 1992Ferrera et al. , 1994. In this study, we also examined labeled neurons in MT, and the preliminary results indicated no correlation between V4 direction strength and labeled neuron numbers in MT (not shown). Thus far, there is no evidence showing that V4 direction neurons obtain direction selectivity from MT.
Since V4 receives inputs from many other visual areas (Ungerleider et al. 2008), all are potential sources for V4 direction selectivity. One study has shown that after V1 lesion, the response of V4 direction neurons is relatively better preserved than that of other types of V4 neurons (Schmid et al. 2013). This suggests that V4 direction-selective neurons may rely heavily on subcortical inputs that by-passing V1. In summary, V4 direction neurons do not seem to rely on inputs from V1, V2, or MT and form a spatialized group in V4 visual processing.
In addition to color, orientation, and direction, it has been shown that V4 has other functional architectures, such as size (Ghose and Ts'o 1997), curvature (Hu et al. 2020;Tang et al. 2020), and binocular disparity (Tanabe et al. 2005;Fang et al. 2019   Correlation between V4 injection sites and V2-labeled neurons a-i. Scatter plot of injection site properties (X) and labeled neuron properties (Y) for all 7 injections (7 dots) in 3 monkeys. Dotted lines represent linear regression of the data points, and gray regions represent 95% con dence intervals. Dots with a cross indicate data from the example cases shown in Figures 1 & 2 (Monkey A). j. Spearman correlation coe cients of the above 9 plots. k. Slope values of the line of t in the above 9 plots.

Figure 4
The V4 color and direction properties of the injection sites were covariant a. Pixel density map that shows pixel numbers (color) having certain color strength (X axis) and orientation strength (Y axis). This map shows the average of data from 12 cases. The color and orientation strengths were calculated in the same way as in Figure 1j. Strong color-selective pixels normally have weak orientation selectivity and vice versa. b. Similar to a but for direction and color strengths. Strong direction pixels tended to have strong color selectivity. c. Similar to a but for direction and orientation strengths. Strong direction pixels tended to have strong orientation selectivity. d. Correlation values for map pairs in a-c. Red crosses are from the example case shown in Figures 1 & 2 (monkey A), blue crosses are for monkeys B and C. **, t-test p < 0.01; ***, p < 0.001. e-g. Scatter plots of different injection site properties for all 7 injections (7 dots). h.
Spearman correlation values for scatter plots in e-g.

Figure 5
Functional properties of V4 injection sites and labeled V2 neurons in monkey C a. A single injection in V4 in monkey C. Scale bar: 2 mm applied to a-b. b. The location of the injection site in 3 functional maps. c.
The functional strength of the V4 injection site. d. Normalized distribution of labeled neurons in 3 types of V2 stripes. e. A single uorescent image of a V2 slice; green color was added for better viewing. Scale bar: 2 mm, applies to e-g. Insert: a zoomed-in view of the framed region, scale bar: 0.1 mm. f. In vivo blood vessel map of area V2 superimposed with labeled neurons (white dots). g. Functional maps for orientation (top) and direction (bottom). h. Distribution of orientation map strength at locations of the labeled neurons. i. Similar to h, but for the direction map.

Figure 6
Summary of functional inputs to V4 from V2 A summary diagram showing the strength of projections from V2 to V4. In V2-V4 projections, V4 color domains mainly receive inputs from V2 thin stripes, whereas V4 orientation domains mainly receive inputs from V2 pale (major) and thick (minor) stripes. Though it is still unclear whether V4 direction domains receive inputs from V2 thin stripes, it is clear that they do not systematically receive inputs from thick or pale stripes.