Eighteen right-handed healthy adult volunteers (8 female, mean age 28.3 ± 9.2 years) took part in the study. None had a history of neurological or psychiatric disorder; written informed consent was obtained from all and the study, which conforms to the Code of Ethics of the World Medical Association, was approved by the Ethics Committee of University College London.

Stimuli were generated using Cogent 2000 and Cogent Graphics (http://www.vislab.ucl.ac.uk/cogent.php) toolboxes running in MATLAB (MathWorks, Na-tick, MA, USA) and were rear projected on a screen by a projector (RM-MSX21G, Victor Company of Japan, Kanagawa, Japan) which has a resolution of 800 × 600 pixels at 60 Hz. Trigger signals were recorded for the MEG system through an IEEE 1284 connection. The delay between the trigger signal and the projection of stimuli (33 ms) was confirmed using a photodiode on the screen before scanning subjects and was corrected during data processing.

To avoid cancelation effects that can occur when both banks of the calcarine sulcus are stimulated (Portin et al., 1999), stimuli were displayed separately in either the lower left or lower right quadrants of the visual field and covered an area of 0° and 7.0° below the fixation cross and 0°–8.4° on either side (Figure 1). Three different categories of stimuli, all composed of the same 26 lines combined into similar forms (six squares and one triangle), were used: faces, houses and ones which could not easily be categorized into either face or house (“Neither”). The latter stimuli were used only to ensure that subjects maintained their attention during the experiment. Two different versions of “House” and “Face” stimuli were used, while the “Neither” category had four versions (see Figure 2).

MEG data were recorded continuously using a 275-channel CTF Omega whole-head gradiometer (VSM MedTech, British Columbia, Canada). Data were sampled at 1200 Hz with a 300 Hz low-pass filter without using high-pass filters. Subjects were fitted with localizer coils at the nasion and at 1 cm anterior to the left and right traguses to monitor head movements during the recording sessions and co-register them to individual MRI structural images acquired in a 3T MRI scanner (Siemens Magnetron Allegra MRI scanner or Trio Tim 3T scanner, Siemens, Erlangen, Germany). Gaze position and blinking were monitored by an EyeLink 1000 eyetracker (SR Research Ltd., Ontario, Canada).

Subjects viewed the display screen at a distance of 60 cm. The experiment consisted of five 5 min runs; each comprised 140 stimuli, 63 of which were “Faces,” 63 “Houses” while 14 belonged to the “Neither” category. The order of stimulus presentation according to category, sub-type and hemifield stimulated (left or right) was randomized. Stimulus presentation lasted 283 ms, with a randomly varied inter-stimulus interval (ISI) of 1500–2000 ms. Subjects focused on the fixation cross throughout the MEG scan and were asked, during the ISI, to indicate which category of stimulus had been presented, by pressing one of 3 buttons using their index, middle, and ring fingers; the ring finger was always used for the “Neither” category, whilst the index and middle fingers were counter-balanced across subjects.

Data were analyzed offline using SPM-8 (Wellcome Trust Centre for Neuroimaging, London, UK; http://www.fil.ion.ucl.ac.uk/spm). They were divided into 1000 ms epochs, each starting 517 ms before stimulus onset. Epochs affected by blink artifacts (detected using the eye-tracker and also by manual inspection of the raw signal data) were discarded and the remaining ones averaged in each condition and baseline corrected. About 140 responses were recorded for each subject, category, and quadrant, except for the “Neither” category (Face and House stimuli presented in left quadrant, 139.7 ± 14.8 ms and 141.0 ± 18.4 ms, respectively; Face and House stimuli presented in right quadrant, 148.2 ± 34.9 ms and 145.1 ± 16.3 ms, respectively; Neither presented in left and right quadrants 35.2 ± 7.6 and 30.4 ± 5.4, respectively). MEG data for the “Neither” category was ignored in the subsequent analyses due to lack of usable epochs. The signal during the 100 ms period preceding stimulus onset was used as a baseline. Software filters produce artifacts (Acunzo et al., 2012; Ramkumar et al., 2013) and are not recommended (VanRullen, 2011); like others before us (Noguchi et al., 2004; Inui et al., 2006; Acunzo et al., 2012), we therefore analyzed our results without filters.

We hypothesized that (1) there is an early component of event-related magnetic fields (ERFs), before 50 ms post-stimulus (ffytche et al., 1995; Inui et al., 2006; Shgihara and Zeki, 2013), and that (2) there would be differences in the amplitude of magnetic responses between Face and House stimuli at the early component of ERFs. To test these hypotheses, we defined a sensor of interest (SOI) which showed the largest root-mean-square (RMS) amplitude of ERF between 25 and 50 ms among the 37 occipital sensors selected by us based on the sensor names MLO 11–53 and MRO 11–53, defined by SPM-8 (SOI approach: Liu et al., 2002; Noguchi et al., 2004) since our previous work had shown an early component of ERF around this time window (27–44 ms) (Shgihara and Zeki, 2013). SOIs were defined for each subject and for each of the 4 conditions [2 forms (Face/House) × 2 quadrants (Left/Right)] separately; the locations of SOIs are shown in the Supplementary Data. We confirmed that there was no difference in the peak time across the four conditions using a Two-Way ANOVA (2 forms × 2 quadrants) with repeated measures (Main effect of form, P = 0.619; Main effect of quadrant, P = 0.107; Interaction, P = 0.690).

Averaged RMS amplitudes at SOIs across 18 subjects showed peaks around 40 ms (37.2–47.2 ms) for all 4 conditions (Figure 4). To confirm that these peaks were significantly larger than the baseline level, we performed an analysis which is schematically represented in Figure 3 (within form comparison): We divided our data into ten 5 ms time windows, covering the period between 0 and 50 ms after stimulus onset. This included the early response period described in previous studies (Shgihara and Zeki, 2013), namely 27–44 ms. We also defined a baseline period between −50 and 0 ms before stimulus onset.

Averaged RMS amplitudes at each time window were compared against baseline. This comparison was carried out for 4 conditions (2 forms × 2 quadrants) separately, using One-Way ANOVA with repeated measures followed by post-hoc t-tests with Ryan correction (Ludbrook, 1991).

To address the second hypothesis (that Face and House stimuli produce responses with different amplitudes), RMS amplitudes at each time window were compared between Face and House stimuli using a paired t-test for each quadrant (left/right) separately (Figure 3: Between forms comparison).

Sensor-level analysis revealed that all four conditions produced early components of ERFs at around 40 ms (37.2–47.2 ms) after stimulus onset, and that Face stimuli produced larger responses than House stimuli for left quadrant stimulation. To learn which brain areas were responsible for producing these responses, we applied source level analyses for the early component of ERFs using SPM-8. Forward modeling was performed between 0 and 50 ms after stimulus onset using a single sphere model (fine mode), and source inversion (estimation) was performed for the peak time window for each condition (shown in Figure 4) using Multiple Sparse Priors (MSP, Greedy Search; Mattout et al., 2005; Friston et al., 2008) for each subject and condition (first level analysis). The source images produced by the first level analysis were smoothed using a Gaussian smoothing kernel of 9 × 9 × 9 mm and taken to the second (between subject) level analysis using t-tests within occipital areas defined by WFU_PickAtlas (http://www.nitrc.org/projects/wfu_pickatlas/). The existence of these ERFs was independently established at a statistically significant level for the sensor-level analysis (Shgihara and Zeki, 2013). Here, we report the source locations of peak level activations at a significance threshold of P(uncorrected) < 0.001 as well as P(FWE corrected) < 0.05. The same source localizations were performed for the other time bins as well. Figure 6 shows source locations at each time bin and for each condition. These sources were accumulated across time bins for each condition using Image Calculator in SPM-8 to locate the areas which were activated at any time during the early component of the response (Figure 7).

Accuracies and response times for subjects' answers were measured during MEG scans (Table 1) and a Two-Way repeated measures ANOVA (quadrant × form) was applied to each analysis. No significant main effect or interaction was found for accuracy [Main effect of Quadrant, F_{(1, 17)} = 1.376, P = 0.257; Form, F_{(1, 17)} = 0.014, P = 0.906; Interaction, F_{(1, 17)} = 4.178, P = 0.057]. Accuracy for each stimulus condition was more than 95%, which shows that the two types of stimuli constituted from the same lines were perceptually accurately, and readily, distinguishable. There were no significant differences between conditions. Although both the main effect of quadrant stimulated and the interaction were not significant, a significant main effect of form was found for response time [Quadrant, F_{(1, 17)} = 1.533, P = 0.233; Main effect of Form, F_{(1, 17)} = 8.057, P = 0.011; Interaction, F_{(1, 17)} = 0.094, P = 0.762]. Differences in form (Face and House) led to significant differences in terms of response time, which was 20 ms longer for House stimuli than Face stimuli.

To better define the early component of ERFs, we calculated a RMS of amplitudes at the SOI which showed the largest amplitude among all 37 occipital sensors, between 25 and 50 ms after stimulus onset. The averaged time course of the RMS amplitudes at the SOIs for the four conditions [2 forms (Face/House) × 2 quadrants (Left/Right)] and across 18 subjects is shown in Figure 4 and a representative RMS time course for a single subject (Subject #12) is shown in Figure S2 in the Supplementary Data. There is, on average, a peak at around 40 ms after stimulus onset for each condition: 37.2 and 39.7 ms for Face and House stimuli presented in the left quadrant, respectively; 47.2 and 44.7 ms for Face and House stimuli presented in the right quadrant, respectively (See Table 2), and the average peak for Face stimuli lasted longer than that for House stimuli for left quadrant presentation. Contour maps at the peak time bins are shown in Figure 5. They show the existence of magnetic sources in occipital areas for all four conditions, although the activation may extend beyond anteriorly.

To confirm these observations, we performed two analyses on the data, divided into 5 ms time windows, as shown in Figure 3: (a) One-Way ANOVA with repeated measures for each of the four conditions (2 forms × 2 quadrants) to confirm that the peak was larger than baseline (within form comparison in Figure 3) and (b) paired t-tests to compare the RMS amplitudes produced by Face and House stimuli at each of the 10 time windows for each quadrant separately (between form comparison in Figure 3) with the following results:

In summary, all four conditions showed an early component of ERF at around 40 ms after stimulus onset and ERF amplitudes elicited by Face stimuli were different (and larger) than those elicited by House stimuli.

We next wanted to determine the cortical source of the regions producing these ERFs, restricting ourselves to determining whether the source was in V1 as well as the (prestriate) areas outside it, without trying to localize the precise source to previously demarcated regions of the prestriate cortex, a difficult task because of the relatively poor spatial resolution of MEG. Source localization was performed using MSP to estimate the sources of the ERFs during the peak time windows (which corresponds in Figure 3, left panel, to the 35–40 ms bin). See also Table 2 and Figure 5). Figures 6, 7 show the sources at different time windows). At the peak time windows, sources were estimated in both striate (V1) and prestriate cortices for all conditions (Figure 5; Table 3 and Table S2); sources in both were significant at peak level P < 0.05 (FWE corrected) for Face and House stimuli presented in the left quadrant. For right quadrant Face stimuli presentation, however, only sources in prestriate cortex were significant while for right quadrant House presentation, only sources in striate cortex were significant.

Sources for other time windows are shown in Figures 6, 7. All four conditions show sources in prestriate cortex which apparently include areas V3, OFA, FFA, PPA, and lateral occipital complex (LOC), although these source locations are not as accurate as one would wish, due to technical limitations (see Discussion).

Sensor-level analysis showed a difference in amplitudes between Face and House stimuli for left quadrant presentation at 25–30 ms after stimulus onset. Although House stimulation did not produce a significant response against the baseline, Face stimuli did so and the sources of the response were localized in both striate and prestriate cortex (Figure 6, Blue square). This shows that the difference in amplitude at the earliest time window is responsible for both striate and prestriate cortex responses.

In summary, Face and House stimuli presented in the left quadrant activated both V1 (striate cortex) and the prestriate cortex outside it with a peak at around 40 ms after stimulus onset, with the activation produced by Face stimuli being larger in amplitude than those produced by House stimuli. For stimuli presented in the right quadrant, activities were less evident at around 40 ms after stimulus onset, in both striate and prestriate cortex. During the whole period of the early component of ERFs, both striate and prestriate cortex were activated for all four conditions in both hemispheres.

The temporal signature, at which a robust response is elicited from cortex in response to face and house stimulation is the N170, which is in the 130–170 time window post-stimulation (Rousselet and Husk, 2008). This means that activity in response to face or house stimulation cannot be distinguished earlier than this time frame, the time up to that period being considered to be taken up by processing of “low-level” features of form (e.g., those of orientation, spatial frequency and contrast, see Johnson and Olshausen (2003) and Rousselet and Husk (2008). 170 ms is considerably longer than the latency with which a response is obtained from V1 and longer than the latencies of the main component of visual perception (i.e., N75, P100, and N145) (see review by Tobimatsu and Celesia, 2006). The difference in latency between the early (N75) response to visual stimulation and the late (N170) component corresponding to the recognition of stimuli as distinct is consistent with a hierarchical strategy in which “low-level” features are processed first. More recently, the latency of activation produced by face stimuli, at least, has been revised downwards. In particular Seeck et al. (1997), using intra- and extra-cranial electrodes, obtained activity at 50 ms post-stimulus when subjects were asked to differentiate between familiar and unfamiliar faces. Other studies using MEG have also shown that face stimuli lead to activation in occipito-temporal cortex before 100 ms (Braeutigam et al., 2001; Meeren et al., 2008) As well, transcranial magnetic stimulation studies (Pitcher et al., 2007) have shown that face perception can be disrupted at 40–50 ms post-stimulus. Although these results might suggest that the cortical processing of faces (and abstract objects) occurs much earlier than previously thought, these evoked response studies used photographs of actual faces as stimuli to elicit responses, which raises the question of whether the early responses may not have been related to the processing of “low-level” features such as lines. Moreover, these studies leave open the question of whether the early responses can be traced to V1 or to prestriate cortex, or both.

In our study, we tried to circumvent these problems by using stimuli constituted from straight lines, thus ensuring that the (early) “low-level” elements constituting the face or house stimuli were uniform throughout, differed in configuration alone and capable of stimulating strongly areas such as V1 and areas of prestriate visual cortex which have high concentrations of OS cells. We also tried to localize the source of the activity that we obtained, to establish whether both V1 and specialized areas of the prestriate cortex react with similar latencies. Our results show that, with both Face and House stimulation, not only is there a response from V1 but also from prestriate cortex, at around 40 ms after stimulus onset. This latency is similar to the earliest latency 28–32 ms in area V5 of prestriate visual cortex obtained after stimulation with fast motion (>22°s⁻¹) (ffytche et al., 1995), 37 ms in V1 (Inui et al., 2006) and 27–44 ms in striate and prestriate cortex (Shgihara and Zeki, 2013), and it is even earlier than the ones reported by previous studies on face perception. Moreover, we have shown that there is a significant difference in the amplitude of the response to faces and houses, suggesting that the two stimuli are differentiated at this early time period. Our results therefore suggest that differentiation between face and house related activity in the visual brain is due to differences in amplitude of the response provoked by the two categories of stimuli rather than to differences in latency.

Our results thus make it plausible to suppose that the brain uses a parallel strategy, in addition to the hierarchical one, to process forms. Although they show that V1 and areas of the prestriate cortex give an early ERF component at around 40 ms, they do not show conclusively the operation of a parallel strategy since it is still conceivable that a signal from V1 originating at, say, 40 ms after stimulus onset would be processed hierarchically by striate and prestriate cortex within the 40–45 ms peak time window. But there are three lines of evidence which suggest strongly that the hierarchical strategy in constructing forms may be supplemented by a parallel one, in which the OS cells of V1 are not the sole source for the construction of forms in the specialized visual areas of the prestriate cortex.

It is well established that the right hemisphere is dominant for face perception (Kanwisher et al., 1997; McCarthy et al., 1997; Pitcher et al., 2007; Yovel et al., 2008). This is probably why sensor level analysis showed a larger RMS amplitude for Face than House in left quadrant presentation, since left quadrant presentations are mainly processed in the dominant (right) hemisphere for face perception. Right hemisphere dominance for face perception in the early response indicates that the cortical processing of face stimuli starts at the earliest stage of visual perception, around 40 ms post-stimulus.

There are five possible problems worth mentioning here:

There are two unexpected results. (1) In the source level analysis, sources were estimated in both left and right striate cortices (Figure 5). This result seems to conflict with the primary rule that striate cortex processes visual information originating from the contralateral visual field. There are two possible explanations for this. Firstly, the estimated source can be distributed to the opposite side of striate cortex, due to the limited spatial resolution of MEG and MSP source estimation (Cicmil et al., 2014). Secondly, our stimuli extended from 8.4 degrees on either side to the vertical meridian. A previous study has shown that there is a strip of central retina which projects to both hemispheres. The width of this strip is 1 to 6 degrees (Marzi et al., 2009), which may account for our results. (2) Sources in striate cortex for right hemifield presentation were not significant at P(FWE) < 0.05, although they were significant at P(unc.) < 0.001. This lack of significant activation (or weak activation) in striate cortex is surprising given that we do have significant prestriate activation. If anything, this result makes the case for parallel processing more emphatic for it suggests the possibility of a more potent input to prestriate than striate cortex with the face and house stimuli that we used. A parallel result may be found in the fact that signals from appropriate fast moving stimuli (>22°s⁻¹) signal reach V5 directly and before they reach V1 (ffytche et al., 1995).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.