Indeed, the inability to decode perceptual content from the PFC runs counter intuitions about PFC functions we have from animal models such as the macaque monkey, where representations of (perceptual) content can be even more directly assessed than with MVPA, by using electrophysiological recordings from single neurons. These studies show that PFC neurons are tuned for and thus represent perceptual features such as visual motion direction (Zaksas & Pasternak 2006) or somatosensory flutter frequency (Romo et al. 1999). Even more direct evidence for the representation of perceptual content in the PFC comes from a recent study by Theofanis Panagiotaropoulos et al. (2012), which shows that single PFC neurons exhibit stimulus-specific activity modulations as a function of subjective perception under flash suppression, a technique that can render visual stimuli temporarily invisible.
In the absence of direct electrophysiological recordings from human PFC, one possible explanation for this discrepancy is that the macaque brain is organized in a totally different way to the human brain. But while theoretically possible, this seems highly unlikely (Passingham 2009; Roelfsema & Treue 2014). Alternatively, one may consider whether certain properties of the neural representations in the PFC may pose limitations to the ability of the fMRI MVPA technique to decode their content. This is in light of the fact that decoding of content from human PFC has been unsuccessful not only in the context of conscious perception, but also in the context of working memory, which has led to a radical reinterpretation of the role of the PFC in this domain (Sreenivasan et al. 2014).
It has been hypothesized that successful decoding of stimulus features such as orientation or motion direction from sensory areas relies upon the presence of orderly spatial arrangements of these features in cortical columns or maps (Freeman et al. 2011; Kamitani & Tong 2005). It is thus worth asking the question whether PFC exhibits a similar map-like structure, or whether the spatial arrangement of features in the PFC already renders the likelihood of decoding any kind of information from its fMRI activity unlikely. For example, while maps representing space have been identified in the human PFC, they are much smaller than retinotopic maps in early visual areas, and intersubject variability in their location is much higher (Hagler & Sereno 2006). Furthermore, it is known from experiments in monkeys that only a subset of the neurons within the PFC subregions in which these maps have been found actually displays any spatial preference (Funahashi et al. 1989; Rainer et al. 1998). Importantly, the PFC also has a more complicated cytoarchitecture than sensory areas, with longer and more complex dendrites that allow for sampling of information from a wider range of inputs (Jacobs et al. 2001), which may affect the spatial scale at which information is represented and can be read out. Nevertheless, the overall picture that arises from studies employing optical imaging and microstimulation in monkeys is that at least several subregions of the PFC are topographically organized in a similar fashion as sensory areas (Roe 2010). However, while the topography of the PFC may be favorable to MVPA, neural representations in the PFC seem to exhibit more complex features and dynamics on a single neuron and population level than the representations in sensory areas where MVPA has been particularly successful. For example, recent studies in monkeys show that PFC representations are very high dimensional (Rigotti et al. 2013), that selectivity is not fixed but can be acquired (Bichot et al. 1996), that selectivity can change over time even within a trial (Stokes et al. 2013), and that populations of PFC neurons represent multiple stimulus dimensions at the same time even if one dimension is unattended (Mante et al. 2013). Thus, the dimensionality and temporal instability of neural representations in the PFC may pose a serious challenge to fMRI MVPA experiments, given that they rely on an inherently slow, hemodynamic signal that integrates neural activity over time.
Putting these and other (Anderson & Oates 2010; Vilarroya 2013) potential limitations of the MVPA approach aside, what other evidence do we have that the PFC is actually involved in conscious perception? In particular, is there causal evidence for a role of the PFC in conscious perception?