Advance preparation in task-switching: converging evidence from behavioral, brain activation, and model-based approaches

Abstract

Recent research has taken advantage of the temporal and spatial resolution of event-related brain potentials (ERPs) and functional magnetic resonance imaging (fMRI) to identify the time course and neural circuitry of preparatory processes required to switch between different tasks. Here we overview some key findings contributing to understanding strategic processes in advance preparation. Findings from these methodologies are compatible with advance preparation conceptualized as a set of processes activated for both switch and repeat trials, but with substantial variability as a function of individual differences and task requirements. We then highlight new approaches that attempt to capitalize on this variability to link behavior and brain activation patterns. One approach examines correlations among behavioral, ERP and fMRI measures. A second "model-based" approach accounts for differences in preparatory processes by estimating quantitative model parameters that reflect latent psychological processes. We argue that integration of behavioral and neuroscientific methodologies is key to understanding the complex nature of advance preparation in task-switching.

Keywords: ERP; advance preparation; evidence accumulation models; fMRI; task-switching.

Figure 1

Cued trials task-switching paradigms include a random sequence of switch and repeat trials with each stimulus preceeded by a cue. Cues signal either the task to be performed on the subsequent target stimulus (task cues) or whether the task will change or repeat upon the subsequent stimulus (transition cues). The task-switching paradigm involves switching between two or more simple tasks defined on the basis of stimulus-set, response-set, or both. Typically, performance is poorer (e.g., slower RT, more errors) on trials that require a change (switch trials) than a repeat task (repeat trials) – a phenomenon commonly referred to as switch cost. Cued trials paradigms include cue-target sequences where a cue carries task-specific (Task A or B) or trial transition (switch or repeat) information prior to target onset and therefore allow presentation of random sequences of switch and repeat. Increasing either cue-target interval (CTI) or response-target interval (RTI) results in a reduction in switch cost, providing evidence for active preparation and passive dissipation processes, respectively (Meiran, Chorev & Sapir, 2000). Many cued trials paradigms use task cueing and map one cue to each task (1:1 cue-to-task mapping) so that a task switch is confounded with a cue change (Logan & Bundesen, 2003). Recent studies use two cues per task (2:1 cue-to-task mapping, as shown above) to separate task switching from cue change effects. We have focused here only on studies using the cued-trials paradigm as it provides tight control over preparation onset on a trial by trial basis.

Figure 2

Functional Magnetic Resonance Imaging (fMRI) studies measure a blood oxygenation-level dependent (BOLD) signal that is related to complex changes in blood flow, blood volume, and oxygenated and deoxygenated haemoglobin (see Logothetis, for a review). fMRI has superior spatial resolution; thus, it provides important information on the neuroanatomy of cognitive processing. (A) Fronto-parietal activity in switch > repeat contrasts in cued task-switching studies, independent of CTI manipulation. Studies were included a) if they included at least one CTI condition that is considered long enough for advance preparation (i.e., CTI>500ms) but not long enough to require maintenance of a prepared state, and b) if they reported Talairach or MNI coordinates for significant Swt-Rpt activity. Hotspots indicate the peak of activity in each cluster reported in each study. Peaks that fall between the presented slices are shown on the closest corresponding slice. It may be seen that while there is variation in the precise localization of switch > repeat activity, there is also a considerable degree of consistency, with DLPFC, VLPFC, PM, pre-SMA and sPPC showing consistent activation. (B) fMRI studies addressing the concept of advance preparation face the challenge of how to disentangle two or more temporally overlapping BOLD responses, i.e., one or more responses that are associated with preparatory activity and one or more responses associated with target-related activity. Although no technique offers a complete solution to this challenge, a number of different approaches have been used in the task-switching literature: a) Early studies used a constant long CTI (e.g. 8s, Kimberg, Aguirre & D'Esposito, 2000) to unambiguously link BOLD activity to cue- and target-related periods. For task switching paradigms, such long intervals might be problematic as the preparatory processes of interest might be masked by other intervening processes that are not under experimental control, such as maintenance and decay of preparatory state (e.g. Rogers & Monsell, ; Jennings & van der Molen, 2005); b) Many studies use a partial trials design, where cue presence and target presence are orthogonally manipulated to precisely disentangle preparation- and target-related activity in cue-only trials (see Ruge, Goschke, & Braver, 2009b). However, target omission can potentially elicit no-go responses (Jamadar et al., 2010a) introducing an unintended manipulation; c) A third approach varies the CTI and extracts BOLD responses associated with cue-related, delay-related, and target-related neural activity by introducing a separate model regressor that is coupled to the length of the CTI (e.g. Bunge, Kahn, Wallis, Miller & Wagner, 2003). To obtain robust estimates, the CTI is typically varied across a relatively wide range (e.g., 4-12 s), which as mentioned above is problematic in task-switching paradigms; d) Purely experimental approaches such as multivariate pattern classification (e.g. Bode & Haynes, 2009) has been used to identify brain regions that appear to express spatial activation patterns specific to certain information that is available during selective time periods during the trial. These new approaches need to be further explored.

Figure 3

Event-related brain potentials (ERP) are extracted from the electroencephalogram (EEG) through a process of signal averaging and filtering. ERPs represent time-varying scalp fields that result from summation of post-synaptic electromagnetic activity generated by neuronal populations in different parts of the brain (Otten & Rugg, 2005). While the location of neuronal generators of ERP activity cannot be directly inferred from scalp recorded activity (Rugg & Coles, 1995), ERPs have excellent temporal resolution providing millisecond accuracy in differentiation between different conditions. Jamadar et al. (2010b) recorded ERP and fMRI using identical paradigms and the same participants on separate occasions. In this cued-trials paradigm (see Figure 1), prepared and unprepared trials were randomly presented with CTI-700ms. Prepared trials included an informative task cue followed by a bivalent target. Unprepared trials included a non-informative cue followed by an informative bivalent target. (A third condition was included on which an informative cue was followed by a target that was not mapped to any response and is discussed in Jamadar et al. (2010b)). (A) RT and RT switch cost were larger on prepared than unprepared trials. (B) Cue-locked ERP waveforms showed an early parietal positivity for prepared vs. unprepared trials, followed by a later parietal positivity for prepared switch vs. prepared repeat trials. Difference waveforms highlight the early positivity for both informative cues and the later positivity for switch cues only. (C) The relationships between ERP and fMRI outcomes (see D) were used to constrain seeded dipole models for ERP source analysis. This approach can identify potential generators of the ERP effect of interest and is the most common approach to multimodal integration (Dale & Halgren, 2001). In this study, seeded dipole analysis revealed that the DLPFC and posterior cingulate were plausible generators of the early positivity for informative cues and the sPPC was a plausible generator for the later positivity for informative switch cues. (D and E) RT-fMRI and ERP-fMRI correlations. RT-fMRI correlations were used to examine neural activity directly related to the preparation and execution of a response. ERP-fMRI correlations allowed the temporal information inherent in the ERP measure to generate hypotheses about the timing of the processes reflected in fMRI activity. For example, these correlations show that DLPFC activation (see D1) is associated with activity occurring earlier in the CTI than sPPC activation (see E1 and E2). This type of distinction would be difficult to achieve using a traditional fMRI approach. Note that ERP-fMRI correlations provide quantitatively different information to a traditional fMRI contrast and can result in different regions of significant activation to more traditional fMRI contrasts. This is because individual variability in the BOLD signal is included in the error variance component of traditional fMRI contrasts whereas correlational analyses reveal important dimensions in BOLD signal variability that are related to inter-subject variation in RT and ERPs. Thus, correlational analyses can be used to infer, albeit indirectly, a relationship between specific cortical regions and sub-processes of the task, but it will only detect processes that show substantial individual differences between subjects (see also Forstmann et al., 2008a,b). On the other hand, regions of activation that are evident in the contrast analyses but are absent from the correlational analyses may reflect either effects that have minimal individual variation or sub-processes of the task that are unrelated to the specific RT and ERP measures that are chosen to correlate. Thus correlations between RT, ERP and fMRI represent an opportunity to examine individual variability in fMRI contrasts and a method for using the temporal information in the ERP signal to examine the temporal dynamics of the fMRI data.

Figure 4

From Ruge et al. (in press). (A) In this cued-trials paradigm, participants (n = 18) performed two blocked task-switching conditions involving either “accuracy feedback” or “effect feedback” after responding. On each trial, the currently relevant task was indicated by a centrally displayed task cue (“H” for horizontal discrimination and “V” for vertical discrimination). (B) Two types of preparatory BOLD activations were identified, associated either with “intentional preparatory control” processes (stronger switch-related activation in the effect-feedback condition than in the accuracy feedback condition) colored in red/yellow or with “attentional preparatory control” processes (similar switch-related activation for both feedback types) colored in blue/pink.

Figure 5

From Karayanidis et al. (2009). (A) Participants (n = 24) used location based task cues to shift between three tasks. Cues could (a) validly predict a task repeat trial (repeat cues), (b) validly predict a shift away from the recently completed task-set and specify the upcoming task (switch-to), (c) validly predict a shift away from the recently completed task-set but leave task identity to target location (switch-away), or (d) be equally likely to be followed by a repeat or switch trial (e.g., non-informative cues leading to either a switch or repeat trial). (B) Switch-away cues resulted in more accurate but not faster responding than non-informative switch cues. Modeling of decision processes showed that this reflected differences in response criterion adjustment. Specifically, both cues that predicted a switch with certainty produced a higher response criterion than cues that predicted a certain (repeat) or likely (non-informative repeat or switch) repeat trial. This indicates trial-by-trial adjustment of response criterion depending on whether the cue provided certainty about an upcoming switch trial. These models also produce a non-decision time parameter that represents processes not directly related to the decision, such as stimulus processing and response execution. In task-switching paradigms, non-decision time is also affected by whether advance preparation has been effectively completed before target onset (see Karayanidis et al., 2009). In contrast to response criterion, non-decision time reduced progressively from non-informative switch to switch-away to switch-to trials, but did not differ between the latter and repeat trials. Hence, cues providing certainty of an upcoming shift away from the current task-set elicited some advance preparation, even when they did not define the new task-set. Interestingly, the decision time advantage of non-informative switch over switch-away trials (resulting from lower response criterion for the former) masked the non-decision time advantage of switch-away over non-informative switch trials (resulting from partial preparation), resulting in no net difference in mean RT. (C) Both switch-to and switch-away cues elicited an early cue-locked positivity that was not seen with either repeat or non-informative cues. (D) The amplitude of early cue-locked positivity was negatively correlated with RT, non-decision time and criterion for switch-to trials and with RT and non-decision time for switch-away trials.