The neural basis of effort valuation: A meta-analysis of functional magnetic resonance imaging studies

Choosing how much effort to expend is a critical for everyday decisions. While effort-based decision-making is altered in common psychopathologies and many neuroimaging studies have been conducted to examine how effort is valued, it remains unclear where the brain processes effort-related costs and integrates them with rewards. Using meta-analyses of combined maps and coordinates of functional magnetic resonance imaging (fMRI) studies (total N = 22), we showed that raw effort demands consistently activated the pre-supplementary motor area (pre-SMA). In contrast, the net value of effortful reward consistently activated regions, such as the ventromedial prefrontal cortex (vmPFC) and ventral striatum (VS), that have been previously implicated in value integration in other cost domains. The opposite activation patterns of the pre-SMA and vmPFC imply a double dissociation of these two regions, in which the pre-SMA is involved in pure effort cost representation and the vmPFC in net value integration. These findings advance our understanding of the neural basis of effort-related valuation and reveal potential brain targets to treat motivation-related disorders.


Introduction 1
Every day, we are faced with choices about whether to invest effort to attain certain goals 2 (Bailey et al., 2016; Salamone et al., 2009). These effort demands are often regarded as costly, such that 3 individuals tend to avoid one action if it requires too much effort (Kool et al., 2010;Kurniawan et al., 4 2010Kurniawan et al., 4 , 2011. The ability to accurately weigh energy requirements against potential benefits (e.g., 5 "effort-based decision-making"), is therefore crucial for optimal goal-directed action, and alterations in 6 this function are believed to be a core component of motivational disorders, such as apathy (Chong and  accumbens, prelimbic/infralimbic cortex (homologous to the vmPFC), or orbitofrontal cortex, reduce the 36 amount of effort rats invested for rewards (Rudebeck et al., 2006;Walton et al., 2009Walton et al., , 2003. 37 Furthermore, neural activity in the ACC, as measured by single unit recordings, varies with cost-benefit 38 weighting (Hillman andBilkey, 2012, 2010)  studies that have independently manipulated net value and decision difficulty showed that these frontal 52 regions, particularly the dorsal ACC, specifically tracked decision difficulty (Hogan et al., 2017;53 induced different patterns of brain activity, making it difficult to judge whether findings from individual 63 studies can be generalized to the cognitive process of interest. A promising approach to address these 64 issues is to quantitatively synthesize fMRI data across multiple studies using an image-based meta-65 analysis (Muller et al., 2018). Relative to traditional meta-analyses based only on peak coordinates of 66 significant activity, an image-based meta-analytic approach uses the full information of the statistical 67 maps from each study, and has greater power to detect small effect sizes (Luijten et al., 2017;Salimi-68 Khorshidi et al., 2009). A previous study showed that even the inclusion of 20% of statistical maps for 69 included studies could significantly improve the precision of a meta-analysis (Radua et al., 2012). 70 Here, we conducted a hybrid coordinate-and image-based fMRI meta-analysis to identify the 71 neural correlates of effort-related cost processing and value integration. Considering their critical roles 72 in response planning, we hypothesized that frontal regions like the ACC, SMA, and AI would be 73 consistently involved in representing prospective effort, independent of the reward offer. We also 74 aimed to test whether effort-related value integration (i.e., the integration of reward value with the 75 effort required to obtain it) relied on the core valuation areas such as the vmPFC and VS or broader 76 frontal regions. 77

Data collection and preparation. 123
We performed two analyses of interest. The first examined activity related to the raw effort 124 involved in the option itself. We included analyses that examined high vs. low effort demands (i.e., 125 categorical contrasts) and those that examined continuous changes in effort (i.e., parametric 126 modulation). The second analysis examined activity related to the prospective net value of an effortful 127 reward. Whenever possible, we used the contrast related to the net value of a single option (i.e., the 128 subjective value of the chosen option discounted by the effort required to obtain it). When this contrast 129 was unavailable, we used the contrast related to the differences between options instead. Studies that 130 only investigated BOLD activity associated with interactions between reward and effort were excluded, 131 as they did not rely on the same discounting assumptions as other measures of net value. It should be 132 noted that one study (Nagase et al., 2018) included two experiments with six common participants, so 133 we selected the experiment with a larger sample size for the meta-analysis. In another study (Chong et 134 al., 2017), all participants took part in both cognitive and physical effort-based decision-making tasks. 135 Thus, we combined the statistical maps from both tasks to avoid selection bias. Finally, one study 136 (Seaman et al., 2018) had a sample that included participants ranging from 22 to 83 years old. However, 137 the authors of this study provided whole-brain maps that controlled for the effect of age, and we chose 138 to include this data in the net value meta-analysis. 139 140

Final Corpus. 141
As shown in Figure 1, 25 studies were ultimately included in the final corpus of studies, which 142 were considered in one or both meta-analyses on raw effort evaluation and effort-reward integration. 143 The raw effort valuation analysis included 15 maps (65%) and 7 coordinates for raw effort processing,

Meta-analysis. 163
Two separate whole-brain meta-analyses were conducted to examine consistent neural 164 correlates of prospective effort and net value processing, respectively. Random-effect models were used 165 to assess the mean effect size of each study, where the weight of a study is the inverse of the sum of its 166 variance and the between-study variance. SDM z-maps were generated by dividing the voxel-wise effect 167 sizes by their standard errors. As these z-values may deviate from a normal distribution, a null-168 distribution was estimated for each meta-analysis from 50 whole-brain permutations. 169 170

Region-of-Interest (ROI) Analysis. 171
We also examined the whole-brain results beyond these a priori ROIs. To reduce the false-181 positive results due to multiple comparisons, we applied a familywise error (

Heterogeneity and Publication Bias. 204
Areas of significant activation were assessed for heterogeneity and publication bias. For each 205 meta-analysis, peaks with heterogeneity l 2 values > 20% were flagged and inspected. In order to assess 206 publication bias, Hedge's g effect size estimates were extracted at the study level for peak voxels of 207 significant clusters. Funnel plots were created and visually inspected. Egger  To directly examine the roles of key regions in raw effort prospect and effort-reward integration, 219 we focused on seven a priori ROIs. Results are summarized in Table 2. The vmPFC showed consistent 220 activations related to net value and deactivations related to prospective effort. The bilateral VS showed 221 a similar activity pattern, but smaller effect sizes for both analyses. In contrast, the pre-SMA showed 222 consistent activations related to effort demand and deactivations related to net value. The ACC and 223 bilateral AI showed similar activity pattern, but smaller effect sizes for both analyses. Figures 2 and 3  224 show the Hedge's g effect sizes for raw effort prospect and net value analyses in the vmPFC and pre-225 SMA ROIs. The forest plots for other regions were shown in Figure S1-S10. 226 227 3.2. Whole-Brain Analysis 228

Prospective Effort 229
We first examined brain regions that were consistently associated with the valuation of 230 prospective effort demands. As illustrated in Figure 4a, the analysis yielded positive effects clustered in 231 the right pre-SMA and adjacent caudal ACC (see Table 3). At a more lenient, uncorrected p < 0.001 232 threshold, other positive foci were detected in the left SMA, right precuneus, and left middle frontal 233 gyrus, and negative foci were detected in the bilateral vmPFC/OFC and left middle temporal gyrus. 234 Heterogeneity I 2 statistics, funnel plots and Egger regressions did not detect excess 235 heterogeneity or publication bias in any significant clusters in the TFCE-corrected findings. However, in 236 the uncorrected analysis, activation in a cluster in the right precuneus was found to be associated with 237 extreme heterogeneity (I 2 =59.50%). 238

3.2.2.Net Value. 240
Next, we examined brain regions that were consistently associated with net value encoding. As 241 illustrated in Figure 4b, the analysis yielded a large cluster connecting cortical and subcortical regions of 242 the medial PFC, VS, dorsal striatum (bilateral putamen and left caudate), and temporal gyrus (see Table  243 3). Analysis also yielded consistent net value activations in a cluster consisting of the bilateral medial and 50.05% respectively, suggesting that findings in these two regions were highly heterogenous. 253 254

Conjunction Analysis 255
Finally, we performed a conjunction analysis to identify areas that are sensitive to both net 256 value and effort requirements. Due to the exploratory nature of this analysis, we used a lenient 257 threshold of uncorrected p < 0.001 at voxel level and k > 20 at cluster level. Note that we used absolute 258 values in the conjunction analysis because of the clearly dissociable effects found in the main 259 prospective effort and net value meta-analyses. We found that the vmPFC and left lateral orbitofrontal 260 cortex were significantly activated by net value but deactivated by effort requirement. The activation 261 pattern was reversed in the pre-SMA and caudal ACC (Figure 4c). However, all of these findings were not 262 detectable after whole-brain TFCE-correction. 263 264

Supplementary analyses 265
To ensure that the results of the net value meta-analysis were not driven by choice difficulty, we 266 reran our analysis excluding four experiments that used the value of two options as their net value 267 metric (e.g. difference in SV of more vs less effortful option). Importantly, the vmPFC and bilateral VS 268 remained to be the foci with highest effect sizes, and the whole-brain activation pattern was 269 qualitatively similar (see Table S1 and Figure S11), suggesting that our main findings were not influenced 270 by the cognitive demands of comparing two options. Moreover, to ensure that our findings were robust 271 when using a broader definition of net value, we also repeated our analysis including two additional 272 studies that used reward and effort interactions as a measure of net value. Main foci and whole-brain 273 activation patterns remained qualitatively similar to the initial net value meta-analysis (see Table S2 and 274 Figure S12). However, deactivations associated with net value were not detected in these 275 supplementary analyses, suggesting that the deactivations in the SMA detected in the main meta-276 analysis were not robust. 277 278

Discussion 279
We conducted a series of combined coordinate-and image-based meta-analyses to examine the 280 neural substrates of effort-based valuation. We first investigated neural activity related to raw effort 281 and net value in seven a priori ROIs previously implicated in value-based decision-making. We found 282 these regions could be broadly divided into two groups that exhibited distinct activity pattern during 283 these two processes, with the vmPFC and pre-SMA as the central node of each. Specifically, the vmPFC 284 was consistently activated during net value integration but deactivated for raw effort representation, 285 whereas the pre-SMA displayed the opposite pattern. The exploratory whole-brain and conjunction 286 analyses further corroborate the ROI analyses. These findings provide strong evidence for a dissociable 287 role of the vmPFC and pre-SMA in the valuation of effort costs, and implicate these two regions as core 288 components of a network that drives motivated behavior. Previous studies have identified effort-related net value signals in other frontal regions, such as 324 the pre-SMA and ACC, which suggests that these regions may be specifically relevant for effort-reward 325 integration. In the current meta-analysis, however, we found that these regions -in particular, the pre-the processing of effort-related costs, rather than value integration per se. These findings align closely 329 with a previous transcranial magnetic stimulation study, in which disruption of the SMA led to decreased 330 effort perception (Zénon et al., 2015). The pre-SMA and dorsal ACC are also recruited to process other 331 types of costs, such as risk (Mohr et al., 2010) and delay (Schüller et al., 2019). A plausible mechanism, 332 therefore, is that these regions serve as a domain-general hub for cost encoding and transfer the cost 333 information to the vmPFC for calculation of net value. Alternatively, neuroeconomic models of effort-334 based decision-making have posited that the ACC, in particular, is involved in good-to-action 335 transformation (Padoa-Schioppa, 2011). Thus, another plausible mechanism is that the vmPFC computes 336 and compares the net value of separate options and passes choice preference to action selection 337 regions, such as the pre-SMA and ACC, for conversion to motor output. 338 Despite strong evidence about the involvement of the caudal ACC, which is close to the pre-339 SMA, in effort costs processing, it should be noted that the ACC, as a whole, is highly heterogeneous 340 The current study has some limitations. First, the sample size of the net value analysis is 353 relatively small. Although the inclusion of statistical images partly offsets this issue, the number of 354 included studies still limited our ability to further explore the effects of potential moderators, such as 355 effort type (i.e., physical vs. cognitive), parameter type (i.e., difference in SV vs. SV of one option), effort 356 execution requirement (i.e., real vs. hypothetical), and reward probability (i.e., cumulative vs. random 357 payout). Because effort-based decision-making is sensitive to reward probability (Barch et al., 2014; features. Second, the majority of the included studies focused on physical effort measured by handgrip 361 devices. These findings should be treated cautiously when generalizing to other formats of effort. 362 Finally, the meta-analytic results reflected consistent regional activations across studies. Although our 363 study identified critical brain regions related to effort-related value integration or cost encoding, how 364 these regions interact with each other to achieve the dynamic valuation process remains to be 365 elucidated by studies using task-based connectivity technique (Hauser et al., 2017) or imaging methods 366 with higher temporal resolution (e.g., magnetoencephalography). 367 In conclusion, this study is the first to use combined image-and coordinate-based meta-analyses 368 to examine neural activity related to effort-related costs and net value. The results showed the pre-SMA 369 is involved in cost representation of prospective effort independent of rewards. In contrast, the vmPFC 370 and VS, which have been implicated in value integration in other cost domains, are also involved in 371 effort-reward integration. These findings further clarify the neural mechanisms underlying effort-related 372 valuation and may provide candidate intervention targets for patients with decreased motivation to 373 exert effort to obtain rewards.