Fine-grained stimulus representations in body selective areas of human occipito-temporal cortex

Highlights

• Body areas primarily distinguish bodies from non-body stimuli.

• Body areas contain information to discriminate also non-preferred stimuli.

• Increasing specificity from EBA to FBA.

• Category representation varies with location within OTC.

• Support for functional homologies between human and monkey body areas.

Abstract

Neurophysiological and functional imaging studies have investigated the representation of animate and inanimate stimulus classes in monkey inferior temporal (IT) and human occipito-temporal cortex (OTC). These studies proposed a distributed representation of stimulus categories across IT and OTC and at the same time highlighted category specific modules for the processing of bodies, faces and objects. Here, we investigated whether the stimulus representation within the extrastriate (EBA) and the fusiform (FBA) body areas differed from the representation across OTC. To address this question, we performed an event-related fMRI experiment, evaluating the pattern of activation elicited by 200 individual stimuli that had already been extensively tested in our earlier monkey imaging and single cell studies (Popivanov et al., 2012, 2014). The set contained achromatic images of headless monkey and human bodies, two sets of man-made objects, monkey and human faces, four-legged mammals, birds, fruits, and sculptures. The fMRI response patterns within EBA and FBA primarily distinguished bodies from non-body stimuli, with subtle differences between the areas. However, despite responding on average stronger to bodies than to other categories, classification performance for preferred and non-preferred categories was comparable. OTC primarily distinguished animate from inanimate stimuli. However, cluster analysis revealed a much more fine-grained representation with several homogeneous clusters consisting entirely of stimuli of individual categories. Overall, our data suggest that category representation varies with location within OTC. Nevertheless, body modules contain information to discriminate also non-preferred stimuli and show an increasing specificity in a posterior to anterior gradient.

Introduction

In our everyday life, we encounter numerous visual stimuli we can easily identify. Neurons sensitive to object properties relevant for identification and categorization have been described in monkey inferior temporal cortex (IT) and are thought to be present in human occipito-temporal cortex (OTC) (Kourtzi and Connor, 2011). Kiani et al. (2007) recorded neural responses across anterior IT cortex to natural and artificial object images. They found that the categorical structure of the objects was represented by the pattern of activity distributed over the cell population. The major distinction was present between animate and inanimate objects; nevertheless, the category of animate objects was further divided into faces and bodies, which were divided further into several finer grained categories like human faces and monkey faces, or human bodies and four-limbed animal bodies. This organization seems to be similar at least at the core structure in human OTC, where a primary animate–inanimate organization with further sub-division of the animate category into faces and bodies has been reported (Caramazza and Shelton, 1998, Cichy et al., 2014, Connolly et al., 2012, Huth et al., 2012, Kriegeskorte et al., 2008). Whereas several studies have investigated the organization of IT/OTC with respect to different categories, much less is known about their representation within individual face and body selective regions. Yet, comparing the representation within category selective regions with the representation across whole IT/OTC might shed new light on the discussion of modular versus distributed coding of object information in general (Haxby et al., 2001, Reddy and Kanwisher, 2006).

Electrophysiological studies reported a high fraction of face-selective cells and strong face category selectivity within fMRI defined face patches (Issa and DiCarlo, 2012, Tsao et al., 2006). The same conclusion was also drawn from fMRI experiments in monkeys and humans focusing on face selective patches in IT and OTC respectively (Liu et al., 2013, Reddy and Kanwisher, 2006). These results seemed to favor a modular organization with respect to faces. However, little is known about the categorical structure within body selective regions. Two single cell studies of the middle (Popivanov et al., 2014) and presumably anterior (Bell et al., 2011) superior temporal sulcus (STS) body patches in monkeys reported that the majority of neurons responded stronger to body compared to non-body stimuli. However, the selectivity for body stimuli was much lower compared to the one reported for faces in the face patches. Also, both studies reported single cells within the body patches that were highly selective for other categories. Consequently cluster analysis in the mid STS body patch showed an initial clustering of body versus non-body stimuli, where the non-body cluster in turn contained two distinct sub-clusters, perfectly separating faces from inanimate objects (Popivanov et al., 2014). How these results translate to the human is presently unknown, because fine grained stimulus clustering in the extrastriate (EBA (Downing et al., 2001)) and the fusiform (FBA, (Peelen and Downing, 2005, Schwarzlose et al., 2005)) body areas has not been investigated till date.

To address this issue, we collected individual human fMRI responses to 200 stimuli showing monkey and human bodies and faces, four-legged mammals, birds, fruits, sculptures and man-made objects. The stimulus set was identical to the one used in our previous monkey imaging and single cell studies (Popivanov et al., 2012, Popivanov et al., 2014). We compared the representational similarity among these stimuli at the fMRI voxel level within body selective regions and also across whole human OTC to investigate the following questions: 1) What is the fine grained categorical structure representing a large number of animate and inanimate objects in EBA and FBA, 2) is there any difference in the representation between body selective regions and whole OTC and finally, 3) can we use the representational structure to address potential homologies between human and monkey body-selective regions?