Chapter 7 Summary and general discussion

Attention allows us to cope with our inability to process all the information that is available to us. Without attention, we would inevitably lose ourselves in the infinitude of present moment sensory input, past memories, and future plans. As such, attention is by definition a limited capacity process—otherwise it would fall prey to the very problem it was meant to solve. But what if there was a way to enhance attention beyond its limited capacity? To be a little bit faster to find what we are looking for, a little more resistant to distraction, a little more vigilant?

In Part II of this thesis, I studied whether attention can be enhanced through non-invasive brain stimulation, using transcranial Direct Current Stimulation (tDCS) in particular. Because attention is a multi-faceted process involving distributed neural networks, I studied both its spatial and its temporal form, and targeted different brain regions.

In Chapter 3, I examined whether tDCS over the frontal eye fields can improve spatial attention. Because the frontal eye fields are primarily involved in the control of eye movements (i.e., overt spatial attention), I used eye tracking to measure the effect of tDCS, while participants made eye movements to sudden onset targets as fast as possible. I predicted that anodal tDCS would increase baseline activity in the frontal eye fields, and would thereby decrease the latency of eye movements. However, eye movement latency during anodal tDCS did not differ from baseline, or from cathodal tDCS, even though a previous study had reported exactly that (Kanai et al., 2012). tDCS also did not affect the accuracy of eye movements.

In Chapter 4 and Chapter 5, I investigated the effects of tDCS on temporal attention, using the attentional blink task. This study built upon earlier work from our group (London & Slagter, 2015), which showed that the effects of tDCS on the attentional blink differ systematically across individuals. Specifically, the effects of anodal and cathodal tDCS were negatively correlated: individuals that benefited from anodal tDCS tended to worsen during cathodal tDCS (or vice versa). In Chapter 4, I first attempted to replicate this result, using a number of different analyses to quantify replication success. All of these suggested that our study was not a successful replication of London & Slagter (2015). In Chapter 5, I examined whether baseline dopamine levels could predict the effects of tDCS on the attentional blink. I measured spontaneous eye blink rate (sEBR) as an index of dopamine, but sEBR was not associated to changes in attentional blink size following tDCS. Attentional blink size and sEBR were also uncorrelated before tDCS onset (in contrast to an earlier study by Colzato et al. (2008)), which probably partly explains the null result.

All the studies on tDCS and attention in this thesis have thus resulted in null findings. In fact, every study from our group that tried to affect attention with transcranial electrical stimulation (tES; including both tDCS and tACS) has produced null results (Schouwenburg et al., 2019; Schouwenburg, Sörensen, Klerk, Reteig, & Slagter, 2018). One of these studies (Schouwenburg et al., 2019) attempted to use tES to counter the decrements in attentional performance I studied in Part III of this thesis. This study built in part on the findings presented in Chapter 6, where reductions in sustained attention were associated with more variability in the phase of theta oscillations over midfrontal electrode sites. However, Schouwenburg et al. (2019) did not find that theta tACS over midfrontal regions reduced the vigilance decrement relatively to a control tACS condition.

The rest of this chapter is focused on discussing these discouraging results. In the next sections, I offer three overlapping categories of explanations: concerning tES and attention specifically, tES studies more generally, and psychological science as a whole. I end with some directions for future research on tES and attention, as well as some general conclusions that we can draw from this thesis and the current state of the field.

7.1 tES and attention

The series of null results in Part II of this thesis is perhaps less surprising when viewed in light of the literature review I presented in Chapter 2. There, I reviewed all published studies (until mid-2016) that used tES to modulate attention; primarily visual search, spatial attention and sustained attention (52 studies in total). In each of these domains, a few studies reported promising outcomes, where tES produced a sizable enhancement of attention. But these were always accompanied by other studies where tES actually worsened task performance, only worked under certain conditions, or had no clear effects at all. It is difficult to get at the source of these differences in outcome, as the studies also varied greatly in their experimental design and choice of stimulation parameters. But in general, we can conclude that enhancements in attention are not easily obtained with tES.

One potential explanation is that the behavioral effects of attention itself are already rather subtle, at least as they are typically measured in the lab. For example, in the Posner task—one of the most widely used paradigms to study attention—participants are cued where a target is likely to appear later, allowing them to shift their attention to this location beforehand. The average benefit that this cue provides (depending on its predictiveness, location, and other parameters) appears to be a decrease in response times to the target of 10–50 ms (Chica et al., 2014). Likewise, attention may also enhance our sensitivity to visual stimuli at a certain location, which according to one study manifests as a 2–8 percentage point increase in contrast (Carrasco, Ling, & Read, 2004). It seems unlikely that tES would be able to further enhance these effects by a large margin. Effects of tES on attention can thus be expected to be small to begin with, which may render it more difficult to obtain them.

There is a different, but related underlying argument here: that the attentional system is functioning close to optimally in healthy individuals. There might simply not be very much room for improvements that tES could bring. In Chapter 3, I suggested this as a possible explanation for our null findings, as the baseline eye movement of our participants was already very fast. One could argue that if task performance is already at ceiling, then cathodal tDCS should still lead to impairments. Yet, the cognitive effects of cathodal tDCS appear to be less consistent (Jacobson et al., 2012).

Moreover, considering that multiple attention-related brain areas will be active at any given moment, stimulation-induced changes in one area could be compensated for by the rest of the network. For instance, spatial attention seems to be governed by a balance in activity between the two hemispheres (Kinsbourne, 1970). Attempts to disrupt this balance by increasing (decreasing) activity in one hemisphere with anodal (cathodal) tDCS could prompt a compensatory response in the other hemisphere.

This inter-hemispheric balance is already disrupted in hemispatial neglect patients, who typically have a lesion in the right hemisphere (Vallar & Perani, 1986). A number of studies have attempted to use tDCS to restore this balance, by applying anodal tDCS to the lesioned hemisphere and/or cathodal tDCS to the unlesioned hemisphere (see Chapter 2). All but one of these reported improvements following tDCS. tES might thus be more effective in clinical samples, where the margins for improvement are larger and network functioning is clearly impaired. Alternatively, tES could also be more effective with repeated applications in multiple sessions (which is typical in clinical studies, to evoke more long-term changes).

7.1.1 tES to enhance sustained attention?

Under some circumstances, the limits of the attentional system become readily apparent, even in otherwise healthy individuals. This was one of the reasons I decided to test whether tDCS can be used to attenuate the attentional blink (Chapters 4 and 5). However, perhaps the limits to attention come most clearly into view when it has to be sustained for a prolonged period of time. In a classic vigilance task—where rare, critical signals have to be discriminated from frequent distractors that do not require a response—performance already starts to decrease within minutes. But we do not yet fully understand why it is so difficult to sustain attention beyond this time span.

In Chapter 6, I examined changes in sustained attention by having participants perform a vigilance task for 80 consecutive minutes, while recording their EEG. I observed the classical vigilance decrement: task performance dropped steadily and reached a low-point after just 20–30 minutes. After 60 minutes, an unexpected motivation boost partially restored task performance, but participants were not able to maintain this level until the end of the experiment. In the EEG, I found that phase clustering of theta-band oscillations was closely associated with these behavioral changes, suggesting that the timing of the neural response to the stimulus became more variable as performance decreased.

The literature review in Chapter 2 also included studies that paired tES with sustained attention tasks. Two studies indeed reported that tDCS prevented performance declines related to time-on-task (Nelson et al., 2014) or sleep deprivation (McIntire et al., 2014). However, two recent experiments from our group (Schouwenburg et al., 2019) were not as successful, despite a much larger sample size. In the first experiment, tDCS over the medial frontal cortex was delivered after 20 minutes of performing the same task I used in Chapter 6. However, neither anodal nor cathodal tDCS was able to stave off the vigilance decrement. Second, partly inspired by the changes in theta-band oscillations I identified in Chapter 6, Schouwenburg et al. (2019) attempted to stimulate the medial frontal cortex with tACS instead. But this approach also did not seem fruitful. If anything, theta tACS appeared to accelerate the vigilance decrement, relative to a control condition with alpha-band stimulation.

These studies differed markedly in tES parameters and experimental design, as was the case for the other studies reviewed in Chapter 2, which makes it difficult to draw overall conclusions. While enhancement of sustained attention could be a promising application of tES, many more studies will be need to determine whether and how this can be done. One complicating factor is that the vigilance decrement itself remains to be fully understood (Fortenbaugh et al., 2017; Hancock, 2013; Johnston et al., 2018). As I also showed in Chapter 6, both motivation and depletion of resources could play a role, as well as other relevant factors that I did not investigate, such as mind wandering or subjective feelings of fatigue. It is not clear which of these processes were affected by tES in the studies that proved successful, nor which of these would be an optimal target for future studies.

7.2 tES challenges

The use of tES to enhance attention might thus be particularly challenging, given the multi-faceted nature of attention, and that we can expect effects to be small in the healthy brain. But the mixed results that I and others obtained probably also stem from fundamental uncertainties about the tES technique. These hold regardless of whether tES is applied in attention research, or in other domains. Many of these have long been known (Bikson et al., 2019; Reato et al., 2019) and must live in the back of the mind of most scientists that use tES. But it may be that—ever since the pioneering studies that successfully applied tDCS to the human motor cortex (Nitsche & Paulus, 2000, 2001)—we have become so inspired that we have taken too great a liberty with the technique, and have not given enough thought to its limitations. I will therefore reiterate four of the most pertinent factors that determine the outcome of tES below⁷.

1. The cellular effects of tES are subtle and complex. The physiological effects of tDCS are usually summarized following the “anodal-excitation / cathodal-inhibition” dichotomy (Jacobson et al., 2012). That is, the effects of tDCS are ascribed to changes in the neural membrane potential, where anodal tDCS depolarizes neurons and thus has an excitatory effect, while cathodal tDCS hyperpolarizes neurons and thus has an inhibitory effect. This simple heuristic is much more complicated in reality, which makes it difficult to predict the overall outcome of tDCS.

First, the effects are highly dependent on the orientation of the electric field. Anodal (cathodal) tDCS is only excitatory (inhibitory) when the polarization is applied at the cortical surface, and the neurons are exactly parallel to the electric field, with the dendrites closest to the electrode. For inversely oriented neurons, the polarization will also be inverted; for tangentially oriented neurons, there will be almost no polarization at all. Because the cortex is highly folded, the orientation of neurons with respect to the scalp surface varies greatly, so applying tDCS at the scalp should always lead to a mix of these three extremes (and all possibilities in between). (Reato et al., 2019)

Second, all of this holds only for the soma, but the net effect of tES is based on the membrane potential in all parts of the neuron (Jackson et al., 2016). Even for a neuron that is perfectly parallel to the electric field, the apical dendrites will be polarized in the opposite direction as the soma (Bikson et al., 2019). This is particularly important for the effects of tES on synaptic plasticity, which could even go in the opposite direction to the online effects (Kronberg, Bridi, Abel, Bikson, & Parra, 2017).

Third, the direct effects of the electric field on membrane polarization are very subtle. Recent studies have measured the electric field that tDCS at 2 mA generates in the human brain, which peaked at 0.5 (Opitz et al., 2016) – 0.8 (Huang et al., 2017) V/m (though note that a lot of studies stimulate at 1 mA instead). Earlier studies estimated the maximum change in the membrane potential to be 0.1 (Bikson et al., 2004) – 0.3 mV per V/m (Radman et al., 2009). So in the best case scenario, tDCS can result in a polarization of 0.05 – 0.25 mV (Bikson et al., 2019). Although tDCS was never presumed to directly elicit action potentials, it is still prudent to realize that a 0.15 mV polarization would amount to only 1% of the change necessary to do so (as a depolarization of at least 15 mV would be needed to go from the resting threshold at -70 mV to the firing threshold at -50 – -55 mV). Vöröslakos et al. (2018) argue that this is simply too weak to elicit reliable effects. They showed that in the living rat brain, stimulation only affected neuronal spiking and membrane potentials at field strengths exceeding 1 V/m. They then measured the electric fields in human cadavers at different intensities of tDCS, and concluded that achieving a field strength of 1 V/m would require as much as 4–6 mA tDCS. Yet others have concluded that the changes in the electric field produced by conventional tDCS still fall within the lower bound of effectiveness (Huang et al., 2017).

All in all, given that the effects of tES on membrane polarization are weak, and that they vary greatly across neurons and neural compartments, there must be more to the immediate effects of tES. Membrane polarization is likely only the initial step in a collection of changes that tES induces in neural circuits, which we are only beginning to understand (Liu et al., 2018). Even less is known about the offline effects of tES, involving synaptic plasticity. More fundamental in vitro and in vivo animal studies are called for to develop a more complete understanding of the neural mechanisms of tES. There is also a need for meso-scale computational models that can simulate the effects of tES on a whole population of neurons (Bestmann et al., 2015; Molaee-Ardekani et al., 2013).

2. The current flow induced by tES is not spatially specific. The precise pattern of current flow is another important determinant of tES outcome. Typically, tES studies are focused on one particular brain area, based on some evidence of its involvement in the cognitive process that the researcher aims to affect. One of the electrodes (in tDCS, usually the anode) is then placed over this area, often based on a scalp position in the 10-20 system, or (more rarely) with MRI-based neuronavigation. However, this does not guarantee that a sufficiently strong electric field is induced in this brain area, nor that this will happen only in this brain area.

For one, a tES montage always consists of two electrodes, so the “reference” electrode also has to be placed somewhere on the body. In the studies in this thesis, I placed it on the forehead, as this is simply what many studies before us did. However, this will lead to opposite polarity stimulation of the brain tissue underneath this electrode. Some opt to circumvent this issue by placing the electrode elsewhere on the body, such as the shoulder. But this increases the inter-electrode distance, which can decrease the size of the effect (Moliadze et al., 2010; Opitz et al., 2015).

Second, the current flow is not restricted to the area under the electrodes, as the simulation in Figure 1.1D already showed. The induced electric field is always more diffuse (Opitz et al., 2015), and may even peak at other locations, such as in between the electrodes (Saturnino, Madsen, Siebner, & Thielscher, 2017). For some montages, the actual pattern of current flow can differ vastly from the intended one (Karabanov et al., 2019). This is especially true if there is an opportunity for the current to shunt through the skin—which may attenuate the current by 60% or more (Vöröslakos et al., 2018)—or other highly conductive tissues, such as cerebrospinal fluid.

Finally, even if a perfectly focal current distribution were achievable, tES can still have more distal effects, as the activity it induces in the target brain area may spread through the network of other areas it is connected with (Knotkova, Nitsche, & Polania, 2019; Wokke et al., 2015).

All of this makes it very difficult to use tES as a tool to localize functions in the brain, or to predict the outcomes of tES according to which brain areas it affects (Karabanov et al., 2019). Researchers should therefore generally try to model the current flow, especially for novel montages, which could show that claims about specific brain areas have to be adjusted—or are not warranted at all. In addition, tES can be combined with neuroimaging techniques to provide more clues as to which brain areas and/or processes were affected (also see the Future directions section for these and other recommendations).

3. The parameter space for tES is vast and largely unexplored. When designing a tES study, one needs to decide on a large number of parameters, which together determine the actual dose that is delivered (Peterchev et al., 2012). These include the stimulation duration (e.g., 10 min, or 30 min), current intensity (e.g., 1 mA, 2 mA, or higher), stimulation waveform (tDCS, tACS, or tRNS), as well as the electrode size and shape (Saturnino, Antunes, & Thielscher, 2015), electrode montage, and many more parameters. There are so many parameters and so many plausible values to set them to, that researchers are faced with a true combinatorial explosion of possibilities.

For the intensity and duration, 1 mA for 20 minutes was the standard for a long time, based on the pioneering motor cortex-tDCS studies (Nitsche & Paulus, 2000, 2001). The problem is that there is no clear reason why these parameters should generalize to other brain areas, given that they have a different neuroanatomical structure, connectivity, and state dynamics (Bestmann & Walsh, 2017). Now, longer and more intense stimulation is becoming more common (e.g. tDCS at 2 mA, or for 30 min) (Bikson et al., 2016; Grossman et al., 2018). However, even in the motor cortex, the canonical effects may not be elicited with these parameters. For example, while 20 minutes of 1 mA anodal tDCS typically increases motor-evoked potentials, one study showed that increasing the duration to 26 minutes leads to a decrease instead (Monte-Silva et al., 2013). Likewise, while 10 or 20 minutes of 1 mA cathodal tDCS has an inhibitory effect on the motor cortex, increasing the current intensity to 2 mA appears to flip the effect to excitation (Batsikadze et al., 2013; Parkin et al., 2018; Samani, Agboada, Jamil, Kuo, & Nitsche, 2019).

The large variability in parameters across studies (as shown in Chapter 2)—and the differences in outcome that they can produce—hamper our ability to integrate across findings. Large-scale studies are necessary that systematically manipulate parameters (e.g. Samani et al., 2019), complemented by efficient ways to optimize them (e.g. Lorenz et al., 2019).

4. The effects of tES are not consistent across individuals. The outcome of tES may also be affected by individual differences in baseline brain state, neuroanatomy, or demographic and other factors (Polanía et al., 2018). In Chapters 4 and 5, I examined individual differences in tDCS effects on the attentional blink, and tried to account for these in terms of baseline cortical excitability and dopamine levels. However, I was not successful on either front. The change in attentional blink size in the anodal tDCS session was not related to the cathodal session, or to baseline spontaneous eye blink rates (a putative measure of dopamine). This is not to say that these factors are not important; we know that baseline brain state can fundamentally change effects of tES. For example, when the motor cortex is not stimulated at rest, but during a cognitive task or motor exercise, the canonical changes in motor-evoked potentials are no longer obtained (Antal, Terney, Poreisz, & Paulus, 2007).

The problem is rather that there are many more factors shaping individual differences in responses to tES, of which baseline cortical excitability or neuromodulator levels may be only a small proportion. Even in motor-cortex tDCS, there is considerable variability in the response—anodal tDCS is not excitatory for everyone, nor is cathodal tDCS inhibitory for everyone (Chew, Ho, & Loo, 2015; Jamil et al., 2017; López-Alonso et al., 2014; Strube et al., 2016; Wiethoff et al., 2014). This may be caused by a diverse array of factors, including gender, age, baseline level of task performance, genetics, hormone levels, smoking behavior, and more (Krause & Cohen Kadosh, 2014; Li et al., 2015). Even differences in head or neural anatomy may determine tES outcome, by causing differences in the pattern of current flow in the brain (Kim et al., 2014; Laakso, Mikkonen, Koyama, Hirata, & Tanaka, 2019). This concern could be alleviated by constructing current flow models for individual participants before the experiment, and adapting the dosage or montage accordingly. Similarly, the influence of baseline cortical excitability can be revealed with neuroimaging, for example through magnetic resonance spectroscopy of GABA and glutamate levels (Filmer et al., 2019; Talsma et al., 2018).

In the above, I discussed four hurdles in the design and interpretation of tES studies: tES effects are subtle and complex, highly dependent on current flow, contingent on the right combination of parameters, and subject to individual differences. Next to these tES-specific factors, inconsistencies in findings across studies may also stem from fundamental issues in current scientific practice, as discussed in more detail next.

7.3 A “crisis of confidence”

Given the substantial challenges involved in tES research, and the many factors that may determine the outcome, the breadth of tES studies that report enhancement effects is remarkable. These cover all aspects of human cognition, such as attention, memory, perception, cognitive control, creativity, arithmetical reasoning, motor learning, and language acquisition (Coffman et al., 2014; Dedoncker et al., 2016; Santarnecchi et al., 2015). The list of successful clinical applications of tES is perhaps even more impressive, including a diverse array of conditions such as chronic pain, aphasia, depression, schizophrenia, epilepsy, dementia, and addiction (Lefaucheur, 2016).

While this string of successes is surely exciting, some have expressed concerns that they simply cannot all be true (Bestmann & Walsh, 2017; Parkin et al., 2015). There is a lingering suspicion in the field that some of these effects must be overstated, or would fail to replicate (Héroux et al., 2017). Medina & Cason (2017) provide some of the most convincing evidence confirming these reservations. They applied a p-curve analysis (Simonsohn, Nelson, & Simmons, 2014) to a random sample of tDCS studies, as well as a collection of tDCS studies on working memory (from a meta-analysis by Mancuso et al., 2016). For any set of studies investigating true effects, the distribution of reported p-values should be significantly right-skewed, i.e., should contain more low p-values (e.g., .01) than higher p-values (e.g. around .05). If the shape of this distribution is different, there is reason to believe the set of studies do not have evidential value. Both of the samples examined by Medina & Cason (2017) lacked evidential value, suggesting that tDCS had no meaningful effect.

These problems are not specific to tES research, as many fields of (social scientific) research have grappled with a lack of evidential value (Brodeur, Lé, Sangnier, & Zylberberg, 2016; Simmons & Simonsohn, 2017) and low rates of replication (Camerer et al., 2018; Klein et al., 2018; Open Science Collaboration, 2015). Particularly in the field of psychology, this realization has sparked a crisis of confidence (also referred to as the “replication crisis”) in many influential findings (Baker, 2015; Pashler & Wagenmakers, 2012). The origin of the crisis can likely be traced back to the use of questionable research practices (John, Loewenstein, & Prelec, 2012), particularly publication bias, hypothesizing after the results are known (HARKing), p-hacking, and low statistical power (Bishop, 2019; Munafò et al., 2017). In the rest of this section, I will discuss these practices in the context of the tES literature.

Publication bias (Rosenthal, 1979) refers to a preference for positive over negative findings, such that studies with null results remain unpublished (“in the file drawer”). This leads to an overrepresentation of positive results (Franco et al., 2014), to such an extent that more than 90% of published studies in psychology and psychiatry support the researcher’s hypothesis (Fanelli, 2012). For any field where a lot of studies are novel and high-risk—which would also include tES—such figures are unlikely to be true. Some meta-analyses of tES studies have indeed uncovered evidence for publication bias (Mancuso et al., 2016). In addition, a recent special issue collected over sixty null results in non-invasive brain stimulation⁸. This includes the study that I report in Chapter 3 (Reteig et al., 2018b), and many other tES studies on attention (Jacoby & Lavidor, 2018; Lanina, Feurra, & Gorbunova, 2018; Learmonth et al., 2017; Schouwenburg et al., 2018; Sheldon & Mathewson, 2018; Tseng, Wang, Lo, & Juan, 2018; Veniero et al., 2017). Such initiatives that encourage researchers to also publish their null tES findings are vital, to ensure the literature accurately reflects the evidence for tES efficacy, and to determine which combinations of parameters do and do not work.

HARKing (Kerr, 1998) and p-hacking (Simmons et al., 2011; Simonsohn et al., 2014) can turn true negatives into false positives. A researcher engages in HARKing (Hypothesizing After the Results are Known) when they adapt their hypothesis to fit the observed results, if the results do not fit their actual hypothesis. When p-hacking, many analyses are performed (implicitly or explicitly), but only the ones that result in a significant p-value are reported. Both p-hacking and HARKing are deceptively easy to commit, and often happen unintentionally. For example, Medina & Cason (2017) may have revealed some indications for p-hacking and/or HARKing in the sample of tDCS studies on working memory (Mancuso et al., 2016). They note that only 5 out of 23 studies reported a significant difference between anodal and sham tDCS, but 20 out of 23 studies reported some significant result, e.g., when adding a covariate, or splitting the sample into sub-types. Both p-hacking and HARKing can be combated by preregistration (Nosek, Ebersole, Dehaven, & Mellor, 2018) of hypotheses and analysis plans. When preregistrations are provisionally accepted for publication and formally reviewed, in the form of a registered report, publication bias is also thwarted (Chambers, Feredoes, Muthukumaraswamy, & Etchells, 2014). Recently, the first registered report in the tES field was published (Boayue et al., 2019), reporting a failure to replicate a study that showed tDCS can increase mind wandering (Axelrod et al., 2015), which I had included in the review on tES and attention (Chapter 2).

Finally, a study is said to have low statistical power when it has a low probability to detect an effect of a specific size. Many tES studies might chase relatively small effects that would require larger sample sizes to reliably detect (Minarik et al., 2016). Results from the analysis by Medina & Cason (2017) suggest that average power might currently be as low as 5–20%. Especially considering that tES effects are subject to individual differences, many studies are likely to be severely underpowered. This could not only lead to a lot of false negative findings, but would also inflate effect sizes for positive findings (Button et al., 2013).

The combined effects of these four (and other) factors can take extreme forms. For example, there are more than 600 published studies on ego depletion: the idea that self-control or willpower is weakened when the pool of limited resources that it draws on is depleted (Inzlicht & Schmeichel, 2012). However, a recent meta-analysis of these studies (Carter, Kofler, Forster, & McCullough, 2015) and a new multi-lab replication study (Hagger et al., 2016) suggest the effect might not exist at all, or is trivially small.

At this point, we cannot escape asking the following question: Could it be that the field of tES got it this wrong? That in our excitement about the potential of tES, we have oversimplified its physiological effects, and have been led astray by questionable research practices? That we have built a house of cards, and it will soon come crashing down?

The results presented in this thesis are certainly not encouraging. But given its limited scope, I cannot really speak to these questions. That said, I think there is still enough reason to be optimistic. When looking at the history of TMS, the field went through similar troubles, but today has matured considerably (Parkin et al., 2015). Also, some tES findings appear to already be beyond doubt, such as the canonical effects on neurophysiology in animal studies, and motor-evoked potentials in humans.

Nonetheless, it is rather humbling that after almost 5000 published studies on tDCS alone⁹, we still feel compelled to ask this question. But it is important to keep asking ourselves this question, as hubris will slow down progress even further. This is clearly demonstrated by the field of candidate gene studies: the endeavor to link genetic polymorphisms (such as the 5-HTTLPR polymorphism for the serotonin transporter gene) to psychological phenotypes (such as depression). Since the first study was published in 2003 (Caspi et al., 2003), around 450 studies on just this one association followed (Border et al., 2019). However, a recent publication found no evidence for this association in samples of 62,000 and upwards, and also showed that all previous studies used sample sizes that were orders of magnitude too low to detect plausible effect sizes (Border et al., 2019). In other words, 16 years worth of research appears to be based on statistical noise—despite the fact that it took only two years for the first non-replication study (Gillespie, Whitfield, Williams, Heath, & Martin, 2005) to appear (Rieckmann, Rapp, & Müller-Nordhorn, 2009). This story unequivocally shows that while science may be self-correcting, this process can be unacceptably slow if we do not pay heed to legitimate concerns that emerge.

7.4 Future directions

The studies in Part II of this thesis and the review in Chapter 2 indicate that the effects of tES on attention are not clear-cut. I discussed three overlapping categories of explanations for the mixed results that characterize the field: concerning tES and attention specifically, tES studies more generally, and psychological science as a whole. In this section, I offer a few recommendations that may hopefully increase confidence in the field and facilitate scientific progress.

Replicate key findings. I have not come across many replications of tES studies; for example, none of the 52 studies I included in the literature review in Chapter 2 were direct replication studies (performed by another research group). It is probable that many studies have figured out a robust stimulation protocol that is replicable. But it also appears likely that many published studies have overestimated effects. Without replication studies, we are not able to weed out the noise from the signal. In Chapters 3, 4, and 5 of this thesis, I have tried and failed to replicate earlier findings. But none of these were set up to be truly decisive. There is a dire need for more direct replications, in the form of registered reports (to prevent p-hacking, HARKing, and publication bias) with larger sample sizes (to combat low statistical power, and interindividual variability) (e.g. Boayue et al., 2019).

Deepen our understanding of tES neurophysiology. This will require a concerted research effort on multiple levels. Everything stands or falls on the low-level neural mechanisms of tES, which need to be further elucidated in animal- or simulation studies. But also at the level of human research, we can take a few steps back and further explore the basic protocols, before taking tES in entirely new directions. In many cases, tES study design and parameter selection is based on conventions, instead of evidence that the chosen protocol is the optimal one. Large-scale studies that systematically explore the parameter space are needed to make more informed choices. For example, Samani et al. (2019) recently probed the effects of motor-cortex tDCS at many different current intensities and stimulation durations. Likewise, there is a new initiative for a multi-center study aiming to more definitively establish the online effect of tACS (Antal et al., 2019).

Add more control conditions: additional tasks and stimulation sites. Both null and positive findings have greater scientific value and are easier to interpret with more tightly controlled experimental designs (Graaf & Sack, 2018; Parkin et al., 2015; Polanía et al., 2018). Control tasks can demonstrate to what extent a putative enhancement is task-specific, and can also uncover whether enhancements in one domain do not come with potential costs in another (Brem et al., 2014a; Iuculano & Cohen Kadosh, 2013). Similarly, control stimulation waveforms or sites can demonstrate how specific a putative enhancement is for a particular stimulation protocol. For example, one shortcoming of all studies in this thesis is that they lacked a sham condition, which makes it hard to discern the effects of anodal/cathodal tDCS from random variation. Note though that some recent studies suggest that participants are not as blind to sham conditions as it originally seemed (Greinacher, Buhôt, Möller, & Learmonth, 2019; Turi et al., 2019). Therefore, applying tES at another location (for which no effects are expected) may provide a better control condition.

Combine tES with neuroimaging. Neuroimaging techniques such as (f)MRI and EEG can both inform and augment tES studies (Bergmann et al., 2016; Thut et al., 2017). First, before data collection, the targeted stimulation site can be localized with (f)MRI scans, to aid precision of electrode placement (as in Chapter 3). Likewise, prior results from neuroimaging studies can inform the choice of stimulation waveform. For example, Schouwenburg et al. (2019) chose their tACS frequency based on the EEG results in Chapter 6, and the target area based on a meta-analysis of fMRI data (Langner & Eickhoff, 2012). Second, neuroimaging data may also be collected during or after application of tES. This can serve to better understand the neural mechanisms of tES-induced changes in behavior, for example by examining changes in neural oscillations following tACS. Similarly, neuroimaging may identify factors that drive individual differences in the behavioral outcome of tES, such as baseline brain state.

Tailor the stimulation dose to individual participants. Some of the inter-individual variability in tES outcome can perhaps be undercut by adapting the montage and stimulation parameters such that everyone receives the same dose. This will require further developments in the computational modelling of current flow. But this field is progressing steadily: the model parameters have been validated using recordings of the electric field in humans (Huang et al., 2017; Opitz et al., 2016), and the analytical pipelines are increasingly user-friendly (Huang, Datta, Bikson, & Parra, 2018; Saturnino et al., 2018).

Design studies with a strong prior on the mechanism. Given the many levels in between the cellular mechanisms and behavioral outcome of tES, the relationships between those levels are often vaguely defined (Bestmann et al., 2015). In Chapter 3, I had at least a rough idea of how tDCS should affect the functioning of the frontal eye fields, and how this should in turn relate to behavioral changes. In contrast, it might be nigh impossible to make a grounded prediction on the effects of dlPFC-tES on moral reasoning—the technique, the area, and the cognitive function are all simply too complex. I struggled more on this front in Chapters 4 and 5, as it is not clear whether and how anodal or cathodal tDCS of the dlPFC should affect the attentional blink.

Test new stimulation protocols that may outperform tES. Some exciting new methods have been developed that may expand the range of non-invasive brain stimulation beyond the current techniques. These include transcranial focused ultrasound (Folloni et al., 2019; Verhagen et al., 2019), temporal interference stimulation (Grossman et al., 2017), and intersectional short pulse stimulation (Vöröslakos et al., 2018). All are capable of more powerful and more focal stimulation than current tES protocols, but also face their own challenges and have not been extensively tested in humans.

7.5 Conclusions

In this thesis, I have mainly explored whether tDCS can be used to enhance attention. A literature review (Chapter 2) revealed that earlier studies reported mixed results. Likewise, the results of the studies that I conducted are not in accord with earlier findings that tDCS may improve spatial (Chapter 3) or temporal (Chapter 4, 5) attention. Finally, sustained attention (Chapter 6) may be an interesting target for enhancement, but a tES study partly based on this work (Schouwenburg et al., 2019) did not prove effective either.

Based on this thesis and general developments in the field, the future of tES to study attention appears uncertain. In principle, tES is a promising and versatile technique: as a scientific method, a tool for enhancement, and a clinical treatment. But its potential in all three of these directions is yet to be fulfilled. The scientific appeal of tES lies in its ability to causally manipulate brain activity, which could ultimately be used to arbitrate between different theories on how cognition arises from brain activity. But at present, this seems to be out of reach. We simply do not know enough about the basic mechanism of tES, and most studies lack methodological rigor. As long as the basic science on tES is inconclusive, it will be difficult to identify optimal protocols for use in cognitive enhancement and clinical applications as well. Conversely, as long as we don’t understand how a particular enhancement or treatment effect comes about, it is of little scientific value (Duecker, Graaf, & Sack, 2014). These concerns have been expressed for years, but still hold true to this day:

When we look at what we have really learned about cognition from tACS, tDCS and tRNS, it is small potatoes. […] Based on the best available studies, from reputable laboratories, we don’t really know where to put the electrodes, we don’t know how robust is the idea that the effects are excitatory or inhibitory, we don’t know what other behaviors are affected, we haven’t tested the methods with real-world tasks and therefore don’t know how they perform outside the lab, and we have no idea in healthy people if they continue to work after more than 2–3 repeated applications.

— Walsh (2013)

This thesis started out with a rather grand introduction to cognitive enhancement (Chapter 1). Some see this future on the horizon already, and point out potential ethical problems in the use of tES for this purpose (Cohen Kadosh, Levy, O’Shea, Shea, & Savulescu, 2012). The potential of tES has also been recognized beyond academia, as it has gained a lot of attention in the media (Dubljević, Saigle, & Racine, 2014). There is even a group of early adopters who have started to use tES at home (Jwa, 2015)—primarily for cognitive enhancement of attention. The interest in tES is further fueled by companies who market tES devices to consumers, along with promises of the stars and the moon (Santarnecchi, Feurra, Galli, Rossi, & Rossi, 2013). These are all important developments that scientists should have a voice in. When we do, we should not forget to be as skeptical towards others as we can be among ourselves (Riggall et al., 2015; Steenbergen et al., 2016; Walsh, 2013; Wurzman et al., 2016). Certainly, the promises of non-invasive brain stimulation are exciting. But it will require a lot of careful research and steady progress to make them a reality.