Real-time neurofeedback training paradigms are designed to teach individuals volitional control over brain states by presenting them with continuously updated metaphors of brain activity and asking them to learn to modulate these representations, often through a process of trial and error. The first neurofeedback studies estimated brain activity with electroencephalography (EEG) and showed not only that humans are capable of gaining volitional control over regionally specific brain activity [e.g., Mulholland et al., 1976
; Schwartz et al., 1976
], but further, that controlling brain activation can be of therapeutic benefit [e.g., Lubar, 1977
; Mills and Solyom, 1974
; Rosenfeld et al., 1996
; Sterman et al., 1975
]. The advent of functional magnetic resonance imaging (fMRI)—with spatial resolution approximating the size of functional neural modules suggested both by cytoarchitectural and electrophysiological data—brought the promise of greater spatial specificity to investigations of the feasibility of neurofeedback training. Indeed, as fMRI data acquisition and analytic techniques developed to allow processing of multiple-voxel fMRI datasets in real time, pioneering studies using realtime fMRI (rtfMRI) documented the controllability of several brain regions; see DeCharms 
and Weiskopf et al. [2004b]
rtfMRI neurofeedback training studies have demonstrated that individuals can learn to control various brain regions of interest (ROIs) with the aid of a neurofeedback signal. Through neurofeedback training, participants have learned to control activity in auditory cortex [Yoo et al., 2006
], sensorimotor cortex [DeCharms et al., 2004
; Yoo and Jolesz, 2002
], supplementary motor area [Weiskopf et al., 2004a
], parahippocampal place area [Weiskopf et al., 2004a
], and dorsal/rostral ACC [DeCharms et al., 2004
; Weiskopf et al., 2003
]. In addition to these areas, which have been implicated primarily in sensorimotor and cognitive processing, investigators have examined the ability of individuals to learn to control key areas involved in emotional experience and regulation, such as the amygdala [Johnston et al., 2010
; Posse et al., 2003
], rostral/ventral ACC [Weiskopf et al., 2003
], and insula [Caria et al., 2007
; Johnston et al., 2010
]. This work is important both because it extends the scope of research examining control of regionally specific brain processes and because it provides insight about the neural mechanisms underlying emotional control.
Posse et al. 
asked participants to generate a sad mood over a 30-s interval and then presented them with a single estimate of their amygdala activity to reinforce their mood induction strategy. While this study demonstrated that amygdala activity can be estimated and its response detected in real time, amygdala neurofeedback was not used to change activity in this structure, but instead, was presented to confirm the effectiveness of participants’ mood regulatory strategy. In an rtfMRI neurofeedback study of the insula, Caria et al. 
presented participants with estimates of brain activity from this structure, updated on a moment-by-moment basis. These researchers demonstrated both that control of the insula increased in a linear manner with neurofeedback training, and that these training effects generalized to a neurofeedback-free post-training scan.
In the first rtfMRI neurofeedback study of a rostral/ventral ACC ROI, Weiskopf et al. 
presented continuously updated indices of activity from dorsal and rostral/ ventral ACC ROIs simultaneously over several training runs. These investigators showed that an exemplary participant became increasingly proficient in increasing activity in both dorsal and rostral/ventral ACC ROIs. Subsequent analyses indicated that learned modulation of the rostral/ ventral ACC ROI occurred in the rostral but not the ventral aspect of the ACC. In a subsequent study from our laboratory, however, we showed that, when presented with an rtfMRI training signal from the subgenual ACC (sACC) specifically, participants could effectively increase and decrease activity in this structure by using, respectively, negative and positive emotion regulatory strategies [Hamilton et al., 2007
In a recent neural model of emotional functioning, Critchley 
posits that the ACC is centrally involved in the generation of affective states. Critchley cites evidence of blunted autonomic response in individuals with dorsal ACC (dACC) lesions, and of dACC hyper-responsivity following circumscribed denervation of the autonomic nervous system, in postulating that the dACC is integral in generating sympathetic autonomic arousal. More speculatively, but supported by a sizable body of research demonstrating that activity in the dorsal and subgenual ACC are negatively intercorrelated [e.g., Pochon et al., 2002
], Critchley posits that the sACC plays a complementary role to the dACC in generating parasympathetic activity. Given this putative involvement of the sACC in emotion generation, it is important to investigate whether control of this structure—and, potentially, the primary emotion processes it subserves—can be learned with the aid of appropriate neurofeedback.
Examining whether individuals can learn to control sACC activity is also important given the consistent association of this structure with various forms of psychopathology. Functional anomalies in the sACC have been implicated in unipolar depression [e.g., Drevets et al., 1997
; Gotlib et al., 2005
; Mayberg et al., 1999
], in bipolar disorder [e.g., Brooks et al., 2006
; Krueger et al., 2000
], and in obsessive-compulsive disorder [e.g., Van Laere et al., 2006
]. Moreover, Mayberg et al. 
demonstrated the antidepressant effect of exogenous down-modulation of sACC activity with microelectrode stimulation, suggesting that endogenous sACC modulation via neurofeedback training may yield similar clinical benefit.
While findings of Hamilton et al. 
provided an important proof of concept regarding the controllability of sACC with emotion-regulatory strategies, the rtfMRI technique used was applied to a small number of participants; thus, the generalizability of these results remains in question. Moreover, participants in that study were given generic seed strategies to regulate sACC activity over a single training run: participants upregulated negative emotion to increase sACC signal and upregulated positive emotion to decrease sACC signal. Thus, the findings from that study do not rule out the possibility that the sACC ROI signal was merely tracking the emotional state of the participant, and that sACC feedback itself did not play an important role in determining sACC activity. Finally, Hamilton et al. did not address whether learned control over the sACC could generalize in the absence of sACC neurofeedback. In addressing these concerns, the present rtfMRI study examines activity in a sACC ROI in an experimental group of participants before, during, and after sACC neurofeedback training, and in a control group run through an isomorphic paradigm using yoked, sham neurofeedback. Thus, this study allows us not only to assess whether sACC neurofeedback facilitates regulation of this structure beyond what may be associated with implementing an emotion-regulatory strategy alone, but further, to determine whether sACC neurofeedback training effects generalize to a neurofeedback-free context. We predicted that participants in the experimental group, but not in the sham neurofeedback control group, would be able to control sACC activity better when shown a sACC neurofeedback signal than when engaging in an emotional regulatory strategy before neurofeedback training. We also predicted that neurofeedback training effects in the experimental group would persist into a neurofeedback-free post-training scan.