Portal de Periódicos da CAPES

Detecting attention‐related rhythms: When is behavior not enough? (Commentary on van der Werf et al. 2021)

2021; Wiley; Volume: 55; Issue: 11-12 Linguagem: Inglês

10.1111/ejn.15322

1460-9568

Ian C. Fiebelkorn,

Neural and Behavioral Psychology Studies

Recent research suggests that environmental sampling is a fundamentally rhythmic process (Benedetto et al., 2019; Fiebelkorn & Kastner, 2019; Schroeder et al., 2010). In primates, both attention-related sampling and exploratory eye movements are seemingly shaped by theta-band activity (3–8 Hz) in the large-scale network that directs both spatial attention and goal-directed eye movements (i.e., the ''attention network''). The Rhythmic Theory of Attention synthesizes such evidence from behavioral and neurophysiological studies to propose that spatial attention is characterized by two alternating attentional states, with the first state promoting attention-related sampling (i.e., sensory functions of the attention network) and the second state promoting attentional shifting and/or eye movements (i.e., motor functions of the attention network; Fiebelkorn & Kastner, 2019). Theta-rhythmic sampling might thus provide a critical balance between attention-related sampling and shifting functions, preventing us from becoming overly focused on any single location (or object) in the environment. In their study, van der Werf et al. (2021) used behavioral data to investigate important questions as to how behavioral relevance and the number of potential target locations influence patterns of theta-rhythmic sampling during spatial attention. Specifically, they examined whether differences in cue validity (i.e., the likelihood that a target will occur at the cued location relative to a non-cued location) change the pattern of rhythmic sampling. It has been hypothesized, for example, that the number of potential target locations could influence the frequency of sampling at each target location (Holcombe & Chen, 2013; Landau & Fries, 2012). That is, an attention-related sampling process that operates at 8 Hz when there is one potential target location might effectively operate at 4 Hz when there are two potential target locations, alternately sampling the two locations (but see Aussel et al. 2021 for another explanation). Regardless of the specific cue validity, however, van der Werf et al. (2021) did not replicate previous evidence of theta-rhythmicity in behavioral performance during a spatial cueing paradigm. Here, I will discuss (i) some potential reasons for these null findings and (ii) some limitations of using exclusively behavioral data to probe rhythmic neural activity. This commentary is not meant to be a criticism of van der Werf et al. (2021), who based their experimental setup on previous work, but rather a discussion of important methodological considerations. Several studies have previously observed the influence of rhythmic neural activity on attention-related sampling in behavioral data (e.g., Fiebelkorn et al., 2018; Fiebelkorn, Saalmann, et al., 2013; Helfrich et al., 2018; Hogendoorn, 2016; Landau & Fries, 2012; Re et al., 2019; Song et al., 2014; VanRullen et al., 2007). These studies have typically demonstrated that the likelihood of detecting a visual target is not fixed during attentional deployment but rather waxes and wanes over time. The precise timing of attention-related peaks and troughs in visual-target detection (see Figure 1a,b) are consistent across species (i.e., humans and monkeys) and between individuals (i.e., between individual monkeys, where we have many trials compiled across multiple recording sessions). Such findings have generated novel insights into the highly dynamic processes through which we explore our environment and sample sensory information (e.g., Benedetto et al., 2019; Fiebelkorn & Kastner, 2019). Although an exclusively behavioral approach to investigating the influence of rhythmic neural activity on attention-related functions has provided novel insights, visualizing the influence of rhythmic neural activity in behavioral data requires a substantial number of trials and relies on some critical assumptions. In this issue, van der Werf et al. (2021) specifically based their study design on the behavioral component of a previous study by Helfrich et al. (2018). The primary purpose of this earlier study was to investigate the neural basis of theta-rhythmic sampling using human electrocorticography (ECoG). Despite low trial counts, Helfrich et al. (2018) also demonstrated significant evidence of theta-rhythmic sampling in their behavioral data. For the study by Helfrich et al. (2018), low trial counts were unavoidable, as the number of trials reflected the realities of collecting data from a patient population in a hospital setting. However, in the van der Werf et al. (2021) study, low trial counts were a consequence of making comparisons across different conditions of cue validity (i.e., the number of trials had to be stretched across three conditions). Below, I will describe some of the reasons why low trial counts should typically be avoided when attempting to visualize the influence of rhythmic neural activity using exclusively behavioral data. When constructing evidence of rhythmic neural activity from behavioral data, each trial contributes a single data point (e.g., either a hit or a miss) within a single temporal window (i.e., within a limited range of cue-target delays). For example, van der Werf et al. (2021) calculated hit rates at different cue-target delays using approximately 12 trials in each temporal window—similar to the number of trials used by Helfrich et al. (2018). Given these low trial counts, any momentary lapse in focus resulting from subject fatigue, for example, has a substantial influence on the behavioral time-series (i.e., hit rate as a function of the time from cue), creating considerable noise in the data. That is, with only 12 trials, each miss lowers the hit rate within a specific temporal window by approximately 8 percentage points. Therefore, a handful of misses attributable to something other than rhythmic neural activity will considerably alter the temporal dynamics of behavioral performance, thereby decreasing the likelihood of detecting the true influence of rhythmic neural activity on behavioral outcomes (e.g., visual-target detection). In addition to including more trials, researchers can help to mitigate some of this noise in their behavioral data by using a self-paced design, where participants initiate each trial and can therefore take breaks as needed (Fiebelkorn, Saalmann, et al., 2013). Noisy data are especially problematic here because an exclusively behavioral approach undoubtedly underestimates the relationship between rhythmic neural activity and behavioral outcomes. A critical source of underestimated effects arises from an important theoretical assumption: it is impossible to visualize evidence of neural oscillations in behavioral data unless there is a consistent phase across trials in the underlying neural oscillations. That is, the underlying oscillatory phases associated with either better or worse behavioral performance have to occur at somewhat consistent time points (i.e., cue-target delays) relative to the presentation of the spatial cue (Figure 1c,d). Theoretically, consistent phase in the underlying neural oscillation results from phase reset associated with a salient stimulus (e.g., a flashing spatial cue). However, neurophysiological studies have repeatedly shown that the phase reset is not perfectly consistent across trials (e.g., see figure S3 from Fiebelkorn et al., 2018). There is, in fact, considerable variability in the phase of underlying neural oscillations following a phase-resetting event. This variability in phase reset decreases the amplitude of the rhythmic effects that are observable in behavioral data (Figure 1c,d). Pairing behavioral data with electrophysiological data, however, can eliminate this reliance on phase reset. The addition of electrophysiological data provides a measure of oscillatory activity (i.e., phase and amplitude) on each trial and at each time point following the presentation of the cue, rather than having to construct evidence of rhythmic neural activity across many trials (as is done for studies that exclusively use behavioral data). Trial-level oscillatory measures from electrophysiological data, such as oscillatory phase, can then be correlated with behavioral data, regardless of whether there is phase consistency across trials (e.g., Busch & VanRullen, 2010; Dugue et al., 2015; Fiebelkorn et al., 2018; Fiebelkorn, Snyder, et al., 2013; Helfrich et al., 2018; Landau et al., 2015). Other variables under researcher control can also influence both effect sizes and the amount of noise in the data. For example, pinning behavioral performance closer to a subject's perceptual threshold avoids ceiling effects that decrease the amplitude of behavioral oscillations. My purpose here is not to discourage behavioral studies designed to investigate the temporal dynamics of cognition or perception. Behavioral data can certainly be enough to make convincing observations about the temporal dynamics of cognitive and perceptual processes. However, we need to carefully consider experimental variables that are under our control, such as trial counts. It may be that electrophysiological data, in addition to behavioral data, are more appropriate for asking nuanced questions that require either multiple conditions (e.g., different levels of cue validity) or a focus on subject-level differences. The data in Figure 1 were collected in the laboratory of Dr. Sabine Kastner. The authors declare no conflict of interest. The peer review history for this article is available at https://publons.com/publon/10.1111/ejn.15322.