Commentary: Attentional control and the self: The Self Attention Network (SAN)

The Self-Attention Network (SAN) model (Humphreys and Sui, 2015) is a recent neurocognitive model to account for self-biases in the allocation of attention. It emerges from psychological, neuropsychological, and neuroimaging evidence on three phenomena: own-name effects, own-face effects, and self-biases in associative matching. Specifically, it posits that our responses to self-related stimuli are differentially subserved by a network comprising three nodes: (i) a general-purpose top-down attentional control network which involves the dorsolateral prefrontal cortex and the intra-parietal sulcus; (ii) a self-representation hub located in the ventromedial prefrontal cortex (vmPFC), and (iii) a bottom-up orientating mechanism which depends on the posterior superior temporal sulcus (pSTS). Accordingly, attentional shifts upon hearing our own name or seeing our own face would rely on interactions among such nodes, mimicking perceptual-saliency effects, and determining emergent behavior. 
 
Though attractive, this proposal features two major caveats. First, the evidence for own-name effects is inconsistent and undermined by psycholinguistic confounds. Second, the node proposed to subserve self-specific information lacks neurofunctional specificity. Here we discuss both issues and advance relevant methodological recommendations. 
 
The model resorts to own-name studies allegedly showing biases for self-related information. However, confirmatory evidence has not been consistently replicated (Yang et al., 2013), especially when own names are compared to other familiar names (e.g., Harris et al., 2004; Kawahara and Yamada, 2004; Tacikowski et al., 2011). Moreover, when they do emerge, own-name effects may be related to non-self-specific psycholinguistic confounds, such as familiarity and frequency. Familiar and frequent words are processed faster (Guttentag and Carroll, 1994) and yield distinctive electrophysiological modulations (Kutas and Federmeier, 2011). Subjectively, own names are typically more frequent and familiar than other proper and common names. Thus, it is difficult to rule out the impact of such variables on the observed effects. 
 
Moreover, there is no neuroanatomical evidence for a dissociation between own names and other proper or familiar names. A review of patients with selective deficits in recalling people's names found no support for a specialized mechanism supporting proper-name—let alone own-name—processing (Hanley and Kay, 1998). Finally, neuroimaging and lesion studies indicate that proper-name processing mainly involves left temporal structures and, less notably, subcortical structures (Semenza, 2006). None of these regions is part of the neuroanatomical components of the SAN model, and there is no rationale for why one's own name should rely on a network separate from that specialized in processing its overarching category. At the very least, the model should specify the relationship between the regions proposed to support own-name biases and the broader networks subserving lexical processing, in general, and proper names, in particular. 
 
The SAN model further posits that self-specific information is subserved by a putative brain region. The model includes “a self-representation hub housed in the ventromedial prefrontal cortex (vmPFC)” (Humphreys and Sui, 2015: 15; emphasis ours). However, this association lacks neurofunctional specificity. The vmPFC is critically engaged by any affective response shaped by conceptual information about specific outcomes (Roy et al., 2012), and by bottom-up processing of external and internal salient stimuli (Cona et al., 2015). Crucially, the vmPFC is key to assess the familiarity of others' faces (Gilboa et al., 2009) and to discriminate them as a function of their relevance (Pegors et al., 2015). Finally, when one's own name is compared with stimuli of similar personal relevance (e.g., the name of a significant other), vmPFC activity is not differentially modulated by self-biased information. So, rather than self-attention in particular, vmPFC activations seem to index increased affective meaning, relevance, or familiarity of self-faces. In brief, the vmPFC does not seem specific enough to constitute a distinct node subserving sui generis self-attention. 
 
A second caveat concerning the role of the vmPFC has been noted by Vallesi (2015). As this author argues, the model assumes strong and mostly unidirectional excitatory connections from the vmPFC to the pSTS. However, this claim is incompatible with lesion data showing that self-bias effects decrease after damage to the former region, but increase following damage to the latter (Sui et al., 2015a). Accordingly, Vallesi (2015) posits that vmPFC activity may be modulated via inhibitory feedback connections from the pSTS, a specification that is not captured by the putative SAN model. 
 
These caveats may be circumvented via methodological innovations. The confounds surrounding own-name research may be avoided through “new nickname” studies. Participants could be given ad hoc nicknames and be referred to by them systematically throughout an experimental session. All names would have the same frequency and familiarity at baseline (namely, zero), and they could be matched for other relevant psycholinguistic variables, such as length or phonological complexity. If an advantage for own nicknames is thus observed, claims for a self-attention bias could be more validly entertained. A conceptually similar paradigm, designed by the very proponents of the SAN model (Sui et al., 2012, 2015b), illustrates the potential usefulness of “new nickname” studies. For example, upon establishing arbitrary associations between geometric shapes and themselves or other people, participants then show reliable self-prioritization effects, independent of psycholinguistic confounds (Sui et al., 2012). This evidence supports the possible benefits of “new-nickname” studies: while these would involve a similar design, they would decrease perceptual-matching demands and more directly address biases in the specific domain of proper-name processing. 
 
Regarding neuroanatomical concerns, functional and structural connectivity analyses would help clarify whether there is a relationship among SAN hubs and regions for selective processing of own names/faces. Moreover, the (relative) specificity for self-information in the vmPFC could be tested by comparing stimuli with similar relevance and familiarity but with different degrees of self-related information. Finally, additional neuropsychological as well as effective and functional connectivity studies could help elucidate the role of excitatory and inhibitory connections between the vmPFC and the pSTS, showing whether the latter structure modulates self-information processing in the former (Vallesi, 2015). Until these limitations are addressed, the SAN model will remain psycholinguistically and neuroanatomically underspecified.

. Cogn. Neurosci. 7, 5-17. doi: 10.1080/17588928.2015 In their Discussion Paper, Humphreys and Sui (2015) review recent data on the relation between self-bias and attention and bring evidence that self-related stimuli, after a simple association, are able to alter the salience of neutral stimuli which is usually a prerogative of monetary and food reward (O'Doherty et al., 2004;Panasiti et al., 2015;Trilla Gros et al., 2015).
The authors review own-name effects, own-face effects and self-biases in associative matching and propose the "Self Attention Network" (SAN), a network model in which dorso-lateral prefrontal cortex (DLPFC) and intra-parietal sulcus (IPS) exert top-down attention-mediated control over left posterior superior temporal sulcus (LpSTS) and ventro-medial prefrontal cortex (vmPFC) which are instead respectively linked to bottom-up orienting of attention and self-related processing.
Here we would like to contribute to the SAN by suggesting that, in addition to considering the behavioral and neural mechanisms that occur when processing supra-modal self-related stimuli, the model would benefit from taking into account the plastic body-centered representation of the self (Maister and Farmer, 2015) and its effects on attention. Specifically, we would like to speculate on how SAN adapts to situations where bodily self-representation is challenged by experimental manipulations that are able to blur self-other distinction, such as shared visuo-tactile stimulation.
Our own face is one of the most important features that define the self and it has a robust representation and a special status in the human cognitive and neural systems (Keenan et al., 2003;Devue and Brédart, 2011). Self-capture and self-advantage effects while processing the self-face, have been extensively found (Tong and Nakayama, 1999;Brédart et al., 2006;Devue et al., 2009). In addition, self-face recognition is linked to the activity of a partially dedicated brain network (Devue and Brédart, 2011) and specific electrophysiological activity (Tacikowski and Nowicka, 2010).
Nevertheless, recent studies show that self-face recognition may be inherently plastic (Tsakiris, 2008;Sforza et al., 2010) as it can be modified by simple visuo-tactile Interpersonal Multisensory Stimulation (IMS) (see Porciello et al., 2016;Sel et al., 2016a for visuo-cardiac IMS). In fact, experiencing tactile stimuli on one's face while seeing synchronous tactile stimulation delivered on the face of another individual, induces changes to self-face representation: a bias in attributing the other's facial features to the self and the illusory experience of looking at oneself in the mirror (Enfacement). At the neural level such IMS modulates the activity of unimodal (inferior occipital gyrus, IOG) and multimodal (right temporo-parietal junction, TPJ and IPS) areas (Apps et al., 2015), both involved in different aspects of self-consciousness: self-location, self-identification, and first person perspective (Blanke, 2012). In particular, we suggested (Bufalari et al., 2015) that TPJ detects the mismatch between self/other tactile sensations while IPS solves the conflict between felt and observed stimuli by integrating multisensory congruent stimuli and remapping the space around the face which ultimately results in an updated self-face representation and in the illusory experience of looking at oneself in the mirror. In line with this view, recent electrophysiological evidence (Sel et al., 2016b) demonstrates that self-specific mismatch detection mechanisms exist in the brain and along with Apps et al. (2015) neuroimaging data, support the idea that self-processing follows predictive coding's principles (Friston, 2009). In such theoretical account multimodal areas update self-representation in order to minimize the surprise generated in unimodal areas by the synchronous IMS (Apps and Tsakiris, 2014). In addition to changing self-face representation, synchronous visuo-tactile IMS is also able to influence basic perceptual processes, such as detection of facial tactile stimuli (Cardini et al., 2013), and importantly higher level attentional mechanisms triggered by self-related face stimuli. Indeed, we recently measured the attentional capture exerted by the self-face and by a friend's face after participants underwent experimental synchronous and control asynchronous IMS (Porciello et al., 2014b). Participants performed a gaze-following task in which they had to look in the direction signaled by an imperative cue while ignoring distracting stimuli, i.e., either the self or a friend's face gazing toward the correct or the wrong direction. After asynchronous IMS, the distracting power of directional gaze is higher when embedded in one's own face than in a friend's face, confirming that selfsimilarity enhances the degree to which gaze orients attention (Hungr and Hunt, 2012). However, after synchronous IMS, namely when other's facial features are assimilated in the self-face representation (Tajadura-Jiménez et al., 2012), the distracting power of self-gaze vanishes and becomes no more distracting than a friend's face. Synchronous IMS can therefore cancel self-gaze attentional capture resulting in the Engazement effect (Porciello et al., 2014b). At the neural level, we hypothesized that Engazement may rely on the interaction between brain circuits involved in self-recognition (i.e., occipito-frontal and parietal regions, Kircher et al., 2001;Platek et al., 2008), multisensory integration and self/other distinction (i.e., TPJ and IPS, Apps et al., 2015) with those involved in reflexive shifts of attention (i.e., dorsal and ventral fronto-parietal networks, Corbetta et al., 2008;Callejas et al., 2014).
In light of the dynamic causal model developed by Humphreys and Sui (2015), we suggest that IMS-induced plasticity of self-face representation may change attentional capture related to selfgaze (i.e., make self-face stimuli less salient) via the activity of ventral (including the specific portion of TPJ connected with the PFC and the insula, Mars et al., 2012) and dorsal fronto-parietal networks which are respectively involved in bottom-up orienting (Corbetta et al., 2008;Mars et al., 2012) and in inhibitory control (Grosbras et al., 2005;Klein et al., 2009;Cazzato et al., 2012;Porciello et al., 2014a) of attention.
In particular, IMS-induced plastic change in self-face representation may reduce the activation of the two ventral SAN's nodes, namely vmPFC, which represents the self and the saliency of self-related stimuli, and posterior STS, which triggers bottom up orienting of attention toward self-related stimuli. Consequently, the dorsal fronto-parietal attentional network, including DLPFC and IPS, has to exert less control to inhibit automatic orienting of attention (e.g., gaze-following behavior) toward stimuli that, after the inclusion of other's features, are no longer coded as self-related.
To sum up, we suggest that the neurocognitive interaction between self-bias phenomena and attention passes through a basic and fundamentally plastic representation of the bodily self that may follow predictive coding rules.

AUTHOR CONTRIBUTIONS
GP, IM-P, and IB have made substantial, direct, and intellectual contribution to the work, wrote the manuscript and approved it for publication.