Superior Neuronal Detection of Snakes and Conspecific Faces in the Macaque Medial Prefrontal Cortex

Snakes and conspecific faces are quickly and efficiently detected in primates. Because the medial prefrontal cortex (mPFC) has been implicated in attentional allocation to biologically relevant stimuli, we hypothesized that it might also be highly responsive to snakes and conspecific faces. In this study, neuronal responses in the monkey mPFC were recorded, while monkeys discriminated 8 categories of visual stimuli. Here, we show that the monkey mPFC neuronal responses to snakes and conspecific faces were unique. First, the ratios of the neurons that responded strongly to snakes and monkey faces were greater than those of the neurons that responded strongly to the other stimuli. Second, mPFC neurons responded stronger and faster to snakes and monkey faces than the other categories of stimuli. Third, neuronal responses to snakes were unaffected by low-pass filtering of the images. Finally, activity patterns of responsive mPFC neurons discriminated snakes from the other stimuli in the second 50 ms period and monkey faces in the third period after stimulus onset. These response features indicate that the mPFC processes fast and coarse visual information of snakes and monkey faces, and support the hypothesis that snakes and social environments have shaped the primate visual system over evolutionary time.


Introduction
The ability to detect and avoid danger in the environment undoubtedly has survival value. Although primates are prey for many kinds of predators including mammalian carnivores and raptors (Isbell 1994), snakes appear to have been the first of the modern predators of primates (Isbell 2006(Isbell , 2009. It has, in fact, been proposed that snakes were the main selective pressure favoring the origin of primate via changes that allowed primates to visually detect nonmoving snakes very quickly, thereby avoiding them before the strike. Behavioral studies have consistently supported this hypothesis. Both nonhuman and human primates, including naïve children who had never seen snakes previously, can visually detect snakes more quickly than other stimuli, even before conscious awareness sets in (Öhman and Mineka 2003;Lobue and DeLoache 2008;Masataka et al. 2010;Soares et al. 2014). Human behavioral and neurophysiological studies also showed enhanced detection of snakes compared with other stimuli (Gomes et al. 2017;Van Strien et al. 2014a,b). In addition to avoiding predators, because most primates are highly social and many often live their entire lives with the same individuals, there must have been strong selective pressure to be able to navigate the social environment effectively (Silk 2007). Not surprisingly, it has been found that faces, especially fearful faces, generate fast, and efficient detection, even in infants (Hershler and Hochstein 2005;Johnson and Fredrickson 2005;Langton et al. 2008;Tamietto and de Gelder 2010;Kawai et al. 2016).
The neural mechanisms allowing rapid visual detection of snakes and emotional faces have been the focus of several recent studies on the subcortical visual system including the superior colliculus, pulvinar, and amygdala (reviewed in Soares et al. 2017). Neurophysiological investigations have recently shown, for instance, that monkey pulvinar neurons respond more quickly and more strongly to snakes and faces than to other stimuli (Le et al. 2013(Le et al. , 2014Nguyen et al. 2013Nguyen et al. , 2014). This visual system has been implicated in fast and coarse visual processing, and it appears to be particularly sensitive to snakes and emotional faces (Morris et al. 1999;Vuilleumier et al. 2003;Le et al. 2013).
The medial pulvinar and amygdala send robust inputs to the medial prefrontal cortex (mPFC) (Porrino et al. 1981;Romanski et al. 1997), a cortical area that has been implicated in the allocation of attention to biologically relevant stimuli (Carretié et al. 2004; Bar et al. 2006). For instance, human fMRI studies show increased activity in the mPFC in response to a virtual predator, live snakes, and emotional faces (Bishop et al. 2004;Mobbs et al. 2007Mobbs et al. , 2009Nili et al. 2010). Previous studies suggest that the PFC is involved in fast and coarse visual processing (Kawasaki et al. 2001; Bar et al. 2006;Chaumon et al. 2014). Given the neural connections from the medial pulvinar and amygdala to the mPFC, and the sensitivity of the mPFC to biologically relevant stimuli, we hypothesize that this cortical area, like the pulvinar, is also involved in fast and coarse visual processing to facilitate detection of snakes and conspecific faces.
In the present study, we tested the salience of snakes and emotional faces as visual stimuli for primates by analyzing neuronal responses to various visual stimuli in the monkey mPFC. To investigate characteristics of information processing in the mPFC, we also tested mPFC neurons with low-and high-pass filtered stimuli. Low-pass filtered stimuli are associated with the subcortical or cortical magnocellular pathway, which provides fast and coarse information, whereas high-pass filtered stimuli are associated with the cortical parvocellular visual system, which provides fine-gained information (Vuilleumier et al. 2003; Bar et al. 2006;Pourtois et al. 2013). We predicted that mPFC neurons would be more sensitive to low-pass filtered stimuli. Here, we report that monkey mPFC neurons responded more rapidly and more strongly to snakes than to other potential predators, and similarly to emotional faces of conspecifics than to neutral faces of conspecifics and faces of humans. We also report that mPFC neurons preferentially responded to low spatial frequency components of those stimuli. Our results suggest that particular ecological and social selective pressures have shaped not only the subcortical nuclei but also the PFC, a cortical area that has greatly expanded in primates.

Subjects and Experimental Setup
Two adult (1 female and 1 male) macaque monkeys (Macaca fuscata), weighing 7.1-8.6 kg, were used in this experiment. Each monkey was individually housed with food available ad libitum. The monkeys were deprived of water in their home cage and received juice as a reward during training and recording sessions. Supplemental water and vegetables were given after each day's session. To assess the monkeys' health, their weight was routinely monitored. The monkeys were treated in strict compliance with the United States Public Health Service Policy on Human Care and Use of Laboratory Animals, the National Institutes of Health Guide for the Care and Use of Laboratory Animals, and the Guidelines for the Care and Use of Laboratory Animals of the University of Toyama. This study has been approved by the Committee for Animal Experiments and Ethics at the University of Toyama. Environmental enrichment (toys) was provided daily, all surgery was performed under anesthesia, and all efforts were made to minimize suffering. The animals were monitored daily to guarantee that their health conditions met the standards of the guidelines indicated above throughout the whole experiment. Figure 1Aa shows the stimulus set, consisting of photos of 8 categories of the stimuli used in the present study; snakes (sn1-4), monkey faces (expressive and neutral faces) (mf1-4), human faces (angry and neutral faces) (hf1-4), monkey hands (mh1-4), nonpredators (np1-4), raptors (ra1-4), carnivores (ca1-4), and simple geometrical figures (circle, cross, square, and star) (si1-4). The monkey faces mf1-2 are nonaffiliative and usually shown in stressful contexts. In the present study, we used the following snake species: snake sn1, cottonmouth snake (Agkistrodon piscivorous) (Wright and Jones 1957), snakes sn2 and sn3, Gloydius tsushimaensis snakes (Isogawa et al. 1994;McDiarmid et al. 1999), and snake sn4, mamushi snake (Gloydius blomhoffii) (McDiarmid et al. 1999;Fukuda et al. 2006). Snakes sn2-4 occur in Japan.

Visual Stimuli
The stimuli were 256 digitized RGB color-scale images with resolution of 227 × 227 pixels. Stimuli were presented on a black background of 0.7 cd/m 2 with their centers at the center of the display. The luminance of each stimulus was determined by measuring luminance of the circular area (radius, 6.35 cm) including each stimulus inside the circle by means of a luminance meter (BM-7A; Topcon, Tokyo). The luminances of these color stimuli were almost identical (6.005-6.445 cd/m 2 ) [luminous intensity (total luminance) ranged from 38.432 to 41.248 mcd]. Luminance of the white areas inside the simple geometric patterns was 36.5 cd/m 2 (total luminance of the circle, cross, square, and star was 45.6, 38.72, 53.592, and 20.64 mcd, respectively). Michelson contrast was not significantly different among the 8 categories of the stimuli [one-way ANOVA, F(7, 24) = 0.861, P > 0.05]. These stimuli were displayed on a CRT monitor with a resolution of 640 × 480 pixels, and the size of the stimulus area was 5-7°× 5-7°.

Transformation of Visual Stimuli
To analyze the features of the stimuli to which the neurons responded, the visual stimuli were transformed (Fig. 1Ab). In quadrate-scrambling, original images were cut into 64 pieces (8 × 8 pieces), and the fragments were randomly reassembled. In spatial filtering, we chose low-pass filter (LPF) with 6 cycles/ image and high-pass filter (HPF) with 20 cycles/image based on previous studies (Vuilleumier et al. 2003;Rotshtein et al. 2007). First, colors of each image were separated into 3 color channels (red, green, and blue), and converted to grayscale images so that both LPF and HPF could be rendered in grayscale. These 3 channel images were then converted into frequency domains by Fourier transformation, and each was processed with Gaussian LPF and HPF. Finally, the 3 channel images were merged again. In phase scrambling, images were scrambled in the Fourier domain (phase-shifted images). This scrambling degrades shape while preserving almost completely the global low-level properties of the original image (i.e., luminance, global contrast, color, spatial frequency amplitude spectrum) (Rossion and Caharel 2011). All images were processed using MatLab 7.0.
In spatial filtering, LPF and HPF stimuli contain information of only limited spatial frequency ranges (either low or high spatial frequencies). Therefore, the LPF and HPF stimuli may differ in their contrast and/or luminance. A statistical analysis of contrast indicated that mean Michelson contrast of the snake stimuli (sn1-4) was smaller in the HPF stimuli than the original, quadrate-scrambled, phase-shifted, and LPF images (P < 0.05, Tukey test after one-way ANOVA). However, the mean difference in contrast between the original and HPF stimuli was very small and negligible (0.025%). In addition, a statistical analysis of total luminance indicated that mean total luminance (sn1-4) was smaller in the LPF and HPF stimuli than the original, quadrate-scrambled, and phase-shifted images (P < 0.001, Tukey test after one-way ANOVA). However, there was no significant difference in total luminance among the original, quadrate-scrambled, and phase-shifted images (P > 0.05, Tukey test after one-way ANOVA).

Behavioral Tasks
The monkey sat in a monkey chair 68 cm away from the center of a 19-in. computer display for behavioral tasks during the training and recording sessions in a shielded room. The CRT monitor was set so that its center was on the same horizontal plane as the monkey's eyes. The monkey chair was equipped with a responding button, which was positioned so that the monkey could easily manipulate it. An infrared charge-coupled device camera for eyemovement monitoring was firmly attached to the chair by a steel rod. During training and recording sessions, the monkey's eye position was monitored with 33 ms time resolution by an eyemonitoring system (Matsuda 1996). A juice reward was accessible to the monkey through a small spout controlled by an electromagnetic valve. A visual stimulus generator (E-Prime 1.0 Professional, Psychology Software Tools, Inc.) controlled the electromagnetic valve and the timing of visual stimuli onset.
The monkeys were trained to perform a sequential delayed nonmatching-to-sample (DNMS) task that required discrimination of the visual stimuli (Fig. 1A). As illustrated in Figure 1B, the task was initiated by a buzzer tone. A fixation cross then appeared in the center of the display. When the monkeys fixated on the cross for 1.5 s within 0.5°-1.0°window, a sample stimulus was presented for 500 ms (sample phase). The control phase was defined as the 100-ms period before the sample phase. After an interval of 1.5 s, the same stimulus appeared again for 500 ms between 1 and 4 times (selected randomly for each trial). Finally, a new stimulus was presented (target phase). When the target appeared, the monkey was required to press a button within 2 s in order to receive a juice reward (0.8 mL) (correct responses). When the monkey failed to respond correctly during the target phase or to press the button before the target phase, the trials were aborted and a 620-Hz buzzer tone was presented. The intertrial intervals (ITIs) lasted 15-25 s (Fig. 1B).
In the DNMS, the monkeys were required to compare a pair of stimuli in each trial (i.e., sample and target stimuli). In the present study, we presented the stimuli randomly both within and between categories. That is, in each trial, 1 of the 32 stimuli was presented as a sample, and 1 of the 32 stimuli that was different from the sample was presented as a target. The 32 stimuli were randomly presented without replacement as samples, that is, each stimulus was presented only once as a sample until all 32 stimuli were presented. A target stimulus was randomly selected from the 32 stimuli except for the stimulus that was shown as a sample.

Training and Surgery
The monkeys were trained to perform the DNMS task for 3 h/ day, 5 days/week. The monkeys reached a 96% correct-response rate after 3 months of training. After completion of this training period, a head-restraining device (a U-shaped plate made of epoxy resin) was attached to the skull under aseptic conditions (Nishijo et al. 1988a,b;Tazumi et al. 2010). The subject was anesthetized with a combination of medetomidine hydrochloride (0.5 mg/kg, i.m.) and ketamine hydrochloride (5 mg/kg, i. m.). The U-shaped plate was anchored with dental acrylic to titanium bolts that were inserted into the skull. We also implanted a reference pin, the location of which was based on the zero coordinates defined in the stereotaxic atlas of the brain of M. fuscata (Kusama and Mabuchi 1970). During the surgery, heart and respiratory functions and rectal temperature were monitored (LifeScope 14; Nihon Kohden Corporation). A blanket heater was used to keep body temperature at 36 ± 0.5°C. Antibiotics were administered topically and systemically for 1 week after the surgery in order to prevent infection. Two weeks after the surgery, the monkeys were retrained while the head was painlessly fixed to the stereotaxic apparatus with the head-restraining device. The performance criterion (>90%) was attained again within 2 weeks.

Neuronal Recording and Data Acquisition
After the monkeys relearned the DNMS task at a >90% correct ratio, we commenced recording neuronal activity. Neuronal activity was recorded from each hemisphere in both subjects. A glass-insulated tungsten microelectrode (0.8-1.5 MΩ at 1 kHz) was stereotaxically inserted into the mPFC vertically to the orbitomeatal plane in a stepwise fashion by a pulse motor-driven manipulator (SM-21; Narishige). Only neuronal activities with a signal-to-noise ratio >3:1 were recorded. The analog signals of neuronal activities, triggers for visual stimuli, juice reward, button pressing, and the X-Y coordinates of eye position were digitized at a 40-kHz sampling rate and stored in a computer via a multichannel acquisition processor (MAP; Plexon) system. The digitized neuronal activities were isolated into single units by their waveform components using the Offline Sorter program (Plexon). Superimposed waveforms of the isolated units were drawn to assess variability throughout the recording sessions and transferred to the NeuroExplorer program (Nex Technologies) for further analysis. If the monkey exhibited signs of fatigue, such as closing the eyes for several seconds or moving the eyes or hands slowly, the experimental session was immediately terminated. In most cases, the unit recording experiment was terminated within 2-3 h. After responses to the 32 visual stimuli were recorded, the transformed images were then presented to the monkeys if single unit activity was still observed.

Analysis of the Basic Characteristics of mPFC Neurons
Spike sorting was performed with the offline sorter program for cluster analysis (Offline sorter, Plexon). Each cluster was checked manually to ensure that the cluster boundaries were well separated and the waveform shapes were consistent with the action potentials. For each isolated cluster, an autocorrelogram was constructed and only units with refractory periods >1.2 ms were used for further analyses. Finally, superimposed waveforms of the isolated units were drawn to check the consistency of the waveforms. Examples of superimposed spike waves of a mPFC neuron and the autocorrelograms of the neuronal spikes are shown in Supplementary Fig. 1. The data indicated that the superimposed waveforms (A) and various waveform parameters in cluster cutting projections (data not shown) were similar across the sessions. The autocorrelograms (B) revealed a refractory period of 2 ms for the mPFC neurons throughout the recording sessions, suggesting that these spikes were recorded from single neurons. These recorded single neurons were classified into putative pyramidal cells and interneurons based on the baseline firing rate and the spike duration, using k-mean clustering (a standard clustering algorithm) in which the number of clusters (2) can be specified (Ison et al. 2011). Z-score standardized mean baseline firing rates and spike widths (peak-to-through distance) were used for the k-mean clustering.
We analyzed single neuronal activity during the following 2 periods: 100 ms before (pre) and 500 ms after (post) the onset of stimulus presentation in the sample phase. Responses in the target phase were not analyzed. The baseline firing rate was defined as the mean firing rate during the 100-ms preperiod. Significant excitatory or inhibitory responses to each stimulus were defined by a Wilcoxon signed-rank test (P < 0.05 for statistical significance) of neuronal activity between the 100-ms preperiod and the 500-ms postperiod. In order to investigate the temporal changes in the neuronal responses, the 500-ms postperiod was divided into ten 50-ms epochs, and the mean neuronal firing rate was calculated for each of these epochs. The response magnitude was defined as the mean firing rate in each epoch minus the mean firing rate during the 100-ms preperiod.
For each neuron, the response magnitudes during the visual stimulation period (for the whole 500-ms period and for each epoch) for all 32 visual stimuli were analyzed by one-way ANOVA (P < 0.05). Response magnitudes between the stimuli were compared by Tukey post hoc tests (P < 0.05). Neurons with a significant main effect were defined as differential neurons.
We also analyzed response latency to visual stimuli. For each neuron, one peri-event histogram was constructed using the entire set of data for all trials and all stimuli. Neuronal response latency was defined as the interval from the onset of stimulus presentation to the time at which the neuronal firing rate exceeded the mean ±2 SD of the baseline firing rate. In addition, for each neuron, individual peri-event histograms were constructed using data for each of the different stimulus categories. We compared the latencies to various stimulus categories to determine whether the characteristics of the specific visual stimuli could modulate the latencies of the mPFC neurons.
All data were expressed as mean ± SEM. Since all available neuronal data were similarly analyzed after recording, no blinding was done. Equality of variances of the data was confirmed before parametric statistical comparisons using SPSS statistics. No statistical methods were used to predetermine sample sizes (number of neurons), but our sample sizes are comparable to those previously reported in the field.

Multidimensional Scaling Analysis
Multidimensional scaling (MDS) is a method that is used to simplify the analysis of relationships that exist within a complex array of data. MDS constructs a geometric representation of the data in order to show the degree of the relationship between stimuli that are represented by the data matrix (see Young (1987) for more details). In the present study, the 32 visual stimuli were used to elicit neural activity in mPFC neurons.
Data matrices of neural activity in a 93 × 32 array derived from the 93 visually responsive neurons were generated. Euclidean distances as dissimilarity between all possible pairs of 2 visual stimuli were calculated by using the visual responses of the 93 mPFC neurons. The MDS program (PROXSCAL procedure, SPSS Statistical Package, version 16) positioned the visual stimuli in the 2D space with the distances between the stimuli representing the original relationships (i.e., Euclidean distances in the present study) (Shepard 1962). Clusters of the visual stimuli were evaluated by discriminant analysis.

Stereotaxic Localization of the mPFC for Recording and Histology
Before recording from the mPFC in each hemisphere, a tungsten marker (diameter: 800 μm) was inserted near the target area under anesthesia, and magnetic resonance imaging (MRI) scans of the monkey head were performed. The locations of mPFC neurons were based on the zero coordinates defined in the stereotaxic atlas of the brain of M. fuscata individuals (Kusama and Mabuchi 1970). After the last recording session, tungsten markers (diameter: 500 μm) were implanted near the target area under anesthesia. Subsequently, the monkeys were deeply anesthetized with an overdose of sodium pentobarbital (100 mg/ kg, i.m.) and perfused transcardially with 0.9% saline followed by 10% buffered formalin. The brains were removed from the skulls and cut into 120-μm sections containing the mPFC. Sections were stained with Cresyl violet. The sites of tungsten markers were determined microscopically. The location of each recording site was then calculated by comparing the stereotaxic coordinates of recording sites with those of marker positions, and was plotted on the actual tissue sections. Locations of visually responsive neurons in the 2 monkeys were compared on the basis of the shapes of the mPFC nuclei, and were replotted on the serial sections of the mPFC of one monkey.
In the present study, the mPFC was divided into 3 areas; an anterior part of the mPFC (the area anterior to the cingulate sulcus), a dorsal part of the mPFC (the area dorsal to the fundus of the cingulate sulcus), and a ventral part of the mPFC (the area ventral to the fundus of the cingulate sulcus) (see Fig. 10B in Results).

Preferential Responses to Snakes and Monkey Faces
Of 538 mPFC neurons recorded, 215 were tested with all 32 visual stimuli in a DNMS task (Fig. 1). Of these 215, 93 neurons responded to one or more stimuli. Classification of 93 responsive neurons based on baseline firing rates and spike widths is shown in Supplementary Fig. 2. Of these 93 neurons, 11 (11.8%) were classified as putative pyramidal neurons, while 82 (88.2%) were classified as interneurons. The 2 populations of neurons showed significant differences in both the baseline firing rates (putative pyramidal neurons: 2.34 ± 1.07 Hz; putative interneurons: 5.15 ± 0.67 Hz; P < 0.05, unpaired t-test) and the spike widths (putative pyramidal neurons: 0.48 ± 0.02 ms; putative interneurons: 0.20 ± 0.01 ms; P < 0.001, unpaired t-test). Figure 2 shows an example of a neuron that responded more strongly to snakes and carnivores (2sn-ca4) than to other stimuli (mf1-si4). Figure 3 shows response magnitudes of this neuron to all the visual stimuli. There was a significant difference among the response magnitudes (one-way ANOVA; F(7, 24) = 197.001, P = 0.001). Post hoc multiple comparisons indicated that mean response magnitudes were significantly larger to the snakes and carnivores than to the other stimuli (Tukey test, P < 0.001). Mean response magnitudes to monkey faces, raptors, and human faces were also significantly larger than those to hands and simple geometrical shapes (Tukey test, P < 0.05).
There were also significant differences in mean response magnitudes to the 8 stimulus categories [repeated measures one-way ANOVA; F(1, 92) = 353.266, P < 0.001] (Fig. 4B). Post hoc multiple comparisons revealed that mean responses were significantly greater to snakes and monkey faces than to nonpredators, simple geometrical shapes, and hands (Bonferroni test, P < 0.05). Differential responses of mPFC neurons cannot be ascribed to luminance variations in this study since all stimuli employed were controlled for luminance and sizes except the simple geometrical shapes (see Materials and Methods). Image scrambling decreased the selective responses to these stimuli (see below), suggesting that these responses were not ascribed to low-level properties of the images, but to the coherent images. Since these response characteristics were very similar to those in the pulvinar neurons (Le et al. 2013) and the mPFC receives afferent projections from the pulvinar (see Introduction), there might be functional relations between the mPFC and pulvinar. Therefore, we analyzed the correlation between the mPFC neuronal responses (present study) and those in the pulvinar (Le et al. 2013), in which 16 of the same visual stimuli (snakes, monkey faces, monkey hands, and simple geometrical figures) were presented in the same task (Fig. 5A). Simple regression analysis of the response magnitudes revealed a significant positive correlation between response magnitudes in the mPFC and those in the pulvinar [r 2 = 0.629, F(1, 14) = 23.741, P < 0.001].
Latencies of mPFC neuronal responses ranged from 75 to 475 ms. There were significant differences in mean response latencies to the 8 stimulus categories (repeated measures oneway ANOVA, P < 0.001) (Fig. 4C). Post hoc multiple comparisons indicated that the mean response latencies were significantly shorter to snakes and monkey faces than to the other stimulus categories (Bonferroni test, P < 0.05). We also analyzed the * * ) ) / Figure 3. Comparison of response magnitudes of the neuron shown in Figure 2 to the 32 visual stimuli. The mean response magnitudes to the snakes and carnivores were significantly larger than those to the other stimuli (*Tukey test after one-way ANOVA, P < 0.05), while mean response magnitudes to monkey faces, raptors and human faces were significantly larger than those to the hands and simple geometrical shapes (*Tukey test after one-way ANOVA, P < 0.05). Error bars indicate SEM. correlation between mean response latencies in the mPFC (present study) and those in the pulvinar (Le et al. 2013) (Fig. 5B). Simple regression analysis revealed a significant positive correlation between response latencies in the mPFC and those in the pulvinar [r 2 = 0.312, F(1, 14) = 6.311, P < 0.05].
The medial PFC has been implicated in auditory or visual communication among conspecifics (Barbas et al. 1999;Amodio and Frith 2006). To investigate this possibility, we further analyzed neuronal responses to monkey and human faces (Fig. 6). In response magnitudes (A), repeated measures two-way ANOVA revealed significant main effects of species (monkey vs.  Figure 7 shows an example of a snake-best neuron tested with the transformed images. This neuron responded strongly to the original snake image (Fig. 7Aa). Low-pass filtering did not affect the neuronal firing to the snake image (Fig. 7Ab) whereas highpass filtering, quadrate-scrambling, and phase-scrambling decreased responses (Fig. 7Ac,d,e). There was a significant difference among the response magnitudes to these stimuli [repeated measures one-way ANOVA; F(3, 45) = 15.623, P < 0.001; Fig. 7B]. Post hoc multiple comparisons indicated that the mean response was significantly greater to the original snake image than to quadrate-scrambling, high-pass filtering, and phase-scrambling snake images (Tukey test, P < 0.001)

Effects of Image Transformation
We tested 27 neurons with all transformed images. Figure  The mean response magnitudes to emotional faces (green columns) were significantly larger than those to neutral faces (blue columns) (**P < 0.01). In addition, the mean response magnitudes to the monkey faces were significantly larger than those to the human faces (a, P < 0.01).
(B) The mean response latencies to emotional faces (green columns) were significantly shorter than those to neutral faces (blue columns) (**P < 0.01). In addition, the mean response latencies to the monkey faces were significantly shorter than those to the human faces (b, P < 0.05). Error bars indicate SEM.
indicated that there were significant main effects of epoch [F(7, 1014) = 59.258, P < 0.001] and stimulus category [F(4, 1014) = 28.614, P < 0.001]. There was also a significant interaction between the 2 factors [F(28, 1014) = 5.208, P < 0.001]. Post hoc multiple comparisons indicated that in epoch 5 (100-125 ms) (Fig. 8B) the mean response magnitudes to the original snake and low-pass filtered images were significantly greater than those to the quadratescrambled, high-pass filtered, and phase-scrambled snake images (Bonferroni test, P < 0.05). In epoch 6 (125-150 ms) (Fig. 8C) the mean response magnitudes to the original snake image were significantly greater than those to quadrate-scrambled, high-pass filtered, and phase-scrambled snake images (Bonferroni test, P < 0.05). Finally, in epoch 7 (150-175 ms) (Fig. 8D), the mean response magnitudes to the original snake image were significantly greater than those to the high-pass filtered images (Bonferroni test, P < 0.05). These results reveal that the low spatial frequency component of the images was important in activating the mPFC neuronal responses to snakes.

Population Coding in the mPFC
The data sets of response magnitudes of the 93 visually responsive mPFC neurons in epochs 1 (0-50 ms), 2 (50-100 ms) and 3 (100-150 ms) after stimulus onset were subjected to MDS analysis (Fig. 9). After measurement of r 2 and stress value for up to 4 dimensions, 2D spaces showed the best results. In the 2D spaces, r 2 values of epochs 1, 2, and 3 were 0.618, 0.803, and 0.869, respectively. In epoch 1, discriminant analyses indicated no significant separation among the stimuli (Table 1). In epoch 2 (Fig. 9A), 2 groups were recognized, a cluster containing the snakes and another containing the remaining stimuli (nonsnakes).
Discriminant analyses indicated a significant separation between the snake and nonsnake stimuli (P = 0.036) ( Table 1). In epoch 3 (Fig. 9B), 3 groups were recognized; a cluster containing the snakes, a cluster containing the monkey faces, and a cluster containing all the other stimuli. Discriminant analyses indicated significant separation among these 3 clusters (P = 0.017) ( Table 1). These results indicated that population activity of the mPFC neurons discriminated snakes and monkey faces. Figure 10A,B shows the locations of electrode penetrations in the mPFC, and recording sites of responsive neurons in the coronal sections, respectively. Figure 11A shows ratios of responsive neurons in the ventral, anterior, and dorsal parts of the mPFC. Ratios of the responsive neurons were significantly greater in the ventral part than in the anterior part (Chi-square test, P < 0.01). Figure 11B shows ratios of snake-and/or monkey face-responsive neurons out of the recorded neurons. Ratios of the snake-and/or monkey face-responsive neurons were significantly greater in the ventral part of the mPFC than in the anterior and dorsal parts of the mPFC (Chi-square test, P < 0.01).

Discussion
In the present work, we found that neurons located mainly in the ventral mPFC responded selectively to snake and monkey face stimuli. The pattern of responses shows that neurons located especially in the anterior cingulate cortex responded selectively to snakes and monkey faces and in ways that facilitate their visual attention: (1) the ratios of neurons that responded best to snakes and monkey faces were larger than those of neurons that responded best to other categories; (2) mean response latencies were faster to snakes and monkey faces than to other stimuli; and (3) population activity of the mPFC neurons discriminated snakes within 100 ms latency. These responses were dependent on low-frequency images; high-pass filtering of the visual stimuli decreased neuronal responses but low-pass filtering did not. Our results provide clear evidence that snakes and monkey faces provide coarse visual information that is effective in eliciting strong and rapid responses from a subset of visually active mPFC neurons.

Responses to Snakes
The main predators of primates are mammalian carnivores, raptors, and snakes (Isbell 1994). Although our previous studies have shown that neurons in the pulvinar respond more rapidly and strongly to snakes than to other stimuli (Le et al. 2013(Le et al. , 2014, we did not compare snakes with other predators. In the current study, we found that mPFC neurons responded more quickly and strongly to snakes than to carnivores and raptors. These results support the Snake Detection Theory, which posits that snakes, as the first of the modern predators of primates, were such powerful selective agents that they favored the origin of primates via modifications of the visual system that enabled rapid detection and thus better avoidance of snakes, modifications that persist even today (Isbell 2006(Isbell , 2009. It has been suggested that the mPFC is involved in sensory processing that directs attention to salient visual stimuli (Dalley et al. 2004;Guillem et al. 2011), including snakes and other predators (see Introduction). Subliminal presentation of threatrelated stimuli captures attention better than does neutral stimuli, and the elevated attention to these stimuli is associated with gray matter volume of the anterior cingulate cortex (Carlson et al. 2012). In the present study, snake-responsive neurons were concentrated in the ventral part of the mPFC that roughly corresponds to the pregenual and subgenual parts of the anterior cingulate cortex. Most of the responsive neurons were putative interneurons in the present study. Interneurons in the mPFC are reported to be highly responsive to sensory cues and implicated in the generation of gamma oscillation and attentional control (Insel and Barnes 2015;Kim et al. 2016). Together, these findings suggest that the responsive neurons might be involved in control of sensory processing and attention during snake presentation. Human fMRI studies reported that the mPFC, especially its ventral part corresponding to the pregenual and subgenual parts of the anterior cingulate cortex, is activated in the initial detection of the potential threat before interaction with the threat occurs (Mobbs et al. 2007(Mobbs et al. , 2009. A behavioral study reported that lesions of pregenual and subgenual parts of the anterior cingulate cortex decreased behavioral responses to snakes in monkeys (Rudebeck et al. 2006). Moreover, the ventral part of the mPFC is implicated in planning of adaptive responses to the threat through its top-down control of the limbic areas (Nagai et al. 2004;Ochsner and Gross 2005;Etkin et al. 2011). These findings suggest that the ventral part of the mPFC might be involved in the planning of adaptive responses to snakes (e.g., avoidance of snakes before a snake strike) by topdown control of the limbic areas through its connections to the amygdala, hypothalamus, and brainstem (Porrino et al. 1981;Price 2005).

Responses to Faces
The present study indicates that facial stimuli had comparable effects on the mPFC neurons to the snake effects in ratios of face-best neurons, response magnitudes, and latencies. These results are consistent with human behavioral studies reporting pop-out effects of faces (Hershler and Hochstein 2005;Langton et al. 2008). However, only conspecific faces elicited fast and  strong responses. Since many primates are highly social with expressive faces, these results suggest mPFC involvement in social cognition. Consistently, human imaging and monkey lesion studies suggest that the mPFC, especially the pregenual part of the anterior cingulate cortex, plays an important role in social cognition and social behaviors in primates (Amodio and Frith 2006;Rudebeck et al. 2006;Noonan et al. 2010). Furthermore, emotion significantly affected responses of the mPFC neurons; responses to emotional faces were larger and faster than those to neutral faces. Human neuropsychological studies reported similar effects of emotional faces in that activation in the brain including the mPFC as well as autonomic responses were enhanced in response to emotional faces compared with neutral faces (Williams et al. 2006;Grabowska et al. 2011), and emotional faces captured attention faster than neutral faces (Öhman et al. 2001;Hahn et al. 2006). In rats, activity of anterior cingulate cortical neurons was increased during competition with conspecifics (Hillman and Bilkey 2012). In mice, the strength of excitatory synaptic inputs to anterior cingulate cortical neurons was correlated with the social hierarchy (Wang et al. 2011). In primates living in social groups, social hierarchies are typically based on agonistic interactions (eg., Isbell and Pruetz 1998), and agonistic interactions often include facial threats (Mandalaywala et al. 2014). These findings suggest that in primates, mPFC neurons responsive to emotional faces might be involved in such threatening situations.

Neural Mechanisms of Snake and Face Detection
The present results also indicated that activity of the mPFC neurons was mediated by low spatial frequency information. Previous studies suggest 2 possibilities (Bar et al., 2006;Pourtois et al. 2013): mPFC might receive low spatial frequency information from either the subcortical pathway including the superior colliculus, pulvinar, and amygdala, or the magnocellular cortico-cortical pathway. Neuropsychological studies reported that patients with V1 lesions displayed enhanced behavioral responses to emotional stimuli presented in the blind visual field compared with neutral stimuli, and that emotional stimuli, presented in the blind visual field, elicited visually evoked potentials comparable to those to the same stimuli presented in the intact visual field, suggesting that the subcortical pathway might provide fast and coarse visual information (de Gelder et al. 1999;Tamietto et al. 2009). This subcortical colliculus-pulvinar-amygdala pathway has been identified by recent tractography studies in both monkeys and humans (Rafal et al. 2015;Diano et al. 2017). Consistently, low spatial frequency facial photos elicited fast field potentials in the human amygdala earlier than cortical responses (Méndez-Bértolo et al. 2016). It has also been reported that enhanced allocation of attention to emotional faces is associated with activity in the anterior cingulate cortex (De Martino et al. 2009;Carlson et al. 2009Carlson et al. , 2012. In a task using backward masking in which conscious recognition of fearful faces was blocked, activity of the anterior cingulate cortex was correlated with that of the amygdala (Carlson et al. 2009). In addition, neuropsychological studies using patients with extinction  phenomenon reported that the PFC, including the mPFC, was activated in response to unconscious presentation of emotional stimuli (Vuilleumier et al. 2002;Tamietto et al. 2015). Finally, results from our study indicated that response magnitudes and latencies to the visual stimuli in the mPFC were significantly correlated with those in the dorsal (medial and lateral) pulvinar (Le et al. 2013). Taken together, these studies suggest that the information from the subcortical pathway plays an important role in attentional allocation in the unconscious stage of visual processing, although the mPFC might receive fast and coarse information from both the subcortical visual and cortico-cortical magnocellular pathways (see earlier discussion). The subcortical visual pathway has been implicated in fast and coarse visual information in low spatial frequency (see Introduction). Previous neurophysiological studies reported that pulvinar neurons selectively process snake information; monkey pulvinar neurons responded more strongly and quickly to snakes than other stimuli, including faces (Le et al. 2013), and population activity of pulvinar neurons discriminated snakes from other stimuli and snake postures (striking vs. nonstriking) in early latencies less than 100 ms (Le et al. 2013(Le et al. , 2014. Moreover, snakes elicited significantly stronger gamma oscillation in the monkey pulvinar (Le et al. 2016). These findings suggest that gamma oscillation in the pulvinar might activate the mPFC in the feed-forward circuits to direct attention to snakes. In addition, the monkey superior collicular, pulvinar, and amygdalar neurons preferentially responded to facial photos as well as face-like patterns Tazumi et al. 2010;Nguyen et al. 2013Nguyen et al. , 2014, and neuronal responses were stronger and faster to emotional faces than to neutral faces in the monkey pulvinar (Le et al. 2013). The similarity of responses to snakes and faces in the mPFC and subcortical pathway also suggest that the mPFC might receive fast and coarse information from the subcortical visual pathway. This might occur via projections from the dorsal pulvinar to the mPFC directly or indirectly through the amygdala (see Introduction). It might also occur via the inferior pulvinar, which receives visual inputs from the retina and superior colliculus, and the inferior pulvinar projects to the amygdala (Diano et al. 2017). The amygdala in turn projects to the mPFC (see Introduction). The present results are consistent with all of these possibilities. Further studies are required to investigate anatomical substrates of mPFC activation by fast and coarse information.

Conclusions
The results presented here indicate that mPFC neurons responded preferentially to snakes and emotional monkey faces compared with other predators, nonpredators, neutral faces, and other visual stimuli. These neurons were mainly located in the pregenual and subgenual parts of the anterior cingulate cortex. Threat from snakes (the Snake Detection Theory) (Isbell 2006(Isbell , 2009 and primate sociality (the Social Brain hypothesis) (Barton and Dunbar 1997;Dunbar 1998) are hypothesized to be important selective forces in brain evolution. Our findings suggest that these selective forces as phyletic memory (Fuster 2009) might have been important in shaping the response characteristics of mPFC neurons.