Neural mechanisms of regulation of empathy and altruism by beliefs of others’ pain

Perceived cues signaling others’ pain induce empathy that in turn motivates altruistic behavior toward those who appear suffering. This perception-emotion-behavior reactivity is the core of human altruism but does not always occur in real life situations. Here, by integrating behavioral and multimodal neuroimaging measures, we investigate neural mechanisms underlying the functional role of beliefs of others’ pain in modulating empathy and altruism. We show evidence that decreasing (or enhancing) beliefs of others’ pain reduces (or increases) subjective estimation of others’ painful emotional states and monetary donations to those who show pain expressions. Moreover, decreasing beliefs of others’ pain attenuates neural responses to perceived cues signaling others’ pain within 200 ms after stimulus onset and modulate neural responses to others’ pain in the frontal cortices and temporoparietal junction. Our findings highlight beliefs of others’ pain as a fundamental cognitive basis of human empathy and altruism and unravel the intermediate neural architecture.


Introduction 39
Aesop's fable 'The boy who cried wolf' tells a story that villagers run or do not run 40 to help a shepherd boy who cries wolf depending on whether they believe that the boy's 41 crying indicates his actual emotion and need. This story illustrates an important 42 character of human altruistic behavior, that is, perceived cues signaling others' suffering 43 drives us to do them a favor only when we believe their suffering is true. Although this 44 important character of human altruism was documented over 2000 years ago in Aesop's 45 fable and is widely observed in current human societies, its psychological and neural 46 underpinnings have not been fully understood. The present study investigated how 47 beliefs of others' pain (BOP) modulate human altruistic behavior independently of 48 perceived cues signaling others' suffering and whether such modulation effects are 49 mediated by changes in empathy for others' pain and relevant brain underpinnings. 50 Empathy engages cognitive and affective processes that support understanding and 51 sharing of other's emotional states and provides a key motivation or a proximate 52 mechanism for altruistic behavior in both humans and other animals (Batson et al., 2015;53 De Waal, 2008; Decety et al., 2016). Empathy can be induced by perceived cues 54 signaling others' pain that activate neural responses in brain regions underlying 55 sensorimotor resonance (e.g., the sensorimotor cortex), affective sharing (e.g., the 56 anterior cingulate cortex (ACC) and anterior insula (AI)), and mental state 57 inference/perspective taking (e.g., the medial prefrontal cortex (mPFC) and 58 temporoparietal junction (TPJ)) ( showing pain inflicted by needle injections into a hand that was believed to be or not to 86 be anesthetized (Lamm et al., 2007). Functional magnetic resonance imaging (fMRI) of 87 brain activity showed weak evidence for modulations of insular responses to perceived 88 pain by beliefs of anesthetization. However, the results cannot be interpreted 89 exclusively by BOP because the stimuli (i.e., needles) used to induce beliefs of numbed 90 and non-numbed hands were different. An ideal paradigm for testing modulations of 91 empathy by BOP independently of perceived cues signaling others' pain should compare 92 brain activities in response to identical stimuli under different beliefs and enable 93 researchers to test how BOP influences altruistic behavior. 94 The current work tested the hypothesis that BOP provides a fundamental cognitive 95 basis of empathy and altruistic behavior by modulating brain activity in response to 96 others' pain. Specifically, we tested predictions that weakening BOP inhibits altruistic 97 behavior by decreasing empathy and its underlying brain activity whereas enhancing 98 BOP may produce opposite effects on empathy and altruistic behavior. In Experiment 1, 99 based on the common belief that patients show pain expressions to manifest their actual 100 emotional state of pain whereas pain expressions performed by actors/actresses do not 101 whereas pain expressions performed by actors/actresses do not indicate an actual painful 144 state. By contrast, reconfirming patient identities enhanced the coupling between 145 perceived pain expressions of faces and the painful emotional states of face owners and 146 thus increased BOP. The participants were then asked to perform the 2 nd _round test in 147 which each photo was presented with patient or actor/actress identity indicated by a 148 word below the photo. The participants had to rate perceived pain intensity of each 149 photo and to report how much they would like to donate from the extra bonus payment 150 to each model. The participants were told that an amount of money would be finally 151 selected from their 2 nd _round donation decisions and presented to a charity organization 152 after the study. Interestingly, to some degree rather than not at all, the participants reported pain and 172 donated to faces with actor/actress identity in the 2 nd _round test, suggesting that 173 weakening BOP did not fully eliminate empathy and altruistic behavior toward those 174 who showed pain expressions.  showed that the rating scores of pain intensity of faces whose identities changed from 192 patient in the 1 st _round test to actor/actress in the 2 nd _round test significantly predicted  Moreover, as a replication, we tested whether changing the participants' intrinsic BOP 221 causes changes in empathy and altruistic behavior in a way similar to that observed in 222 Experiment 1. In addition, we assessed whether changing intrinsic BOP modulated 223 sharing of others' painanother key component of empathy (Jackson et al., 2005), 224 and whether BOP induced emotional sharing mediates the relationship between BOP 225 and altruistic behavior. 226 To address these issues, we tested an independent sample (N = 60) using the stimuli 227 and procedure that were the same as those in Experiment 1 except the following. In the 228 1 st _round test the participants were informed that they were to be shown with photos 229 with pain expressions taken from patients who suffered from a disease and received a 230 medical treatment. After the presentation of each photo the participants were asked to 231 estimate, based on perceived pain expression of each face, how effective the medical 232 treatment was for each patient by rating on a Likert-type scale (0 = no effect or 0% 233 effective, 100 = fully effective or 100% effective). The rating scores were used to 234 estimate the participants' intrinsic BOP of each face with a higher rating score 235 (indicating more effective treatment) corresponding to a weaker BOP as a more 236 effective medical treatment reduces a patient's pain to a greater degree. In addition to 237 rating pain intensity of each face, the participants were asked to report how unpleasant 238 they were feeling when viewing each photo by rating on a Likert-type scale (0 = not 239 unpleasant at all, 10 = extremely unpleasant). The unpleasantness rating was performed 240 to assess emotional sharing of others' pain (Jackson et al., 2005). In the 2 nd _round test 241 the participants were told that the medical treatment was actually fully effective for half 242 patients but had no effect for the others. Each photo was then presented again with 243 information that the medical treatment applied to the patient was 100% effective (to 244 decrease the participants' beliefs of the patients' painful states) or 0% effective (to 245 enhance the participants' beliefs of the patients' painful states). Thereafter, the 246 participants were asked to perform the rating tasks and to make monetary donation 247 decisions, similar to those in the 1 st _round test. 248 To assess whether individuals' intrinsic BOP predicted their empathy and altruistic 249 behavior across different target faces, we conducted Pearson correlation analyses of the 250 relationships between intrinsic BOP as indexed by the rating score of treatment 251 effectiveness and empathy rating scores/amounts of monetary donations across the 252 sixteen models in the 1st_round test in each participant. The correlation coefficients 253 were then transformed to Fisher's z values that were further compared with zero. 254 One-sample t-tests revealed that the z values were significantly smaller than zero 255  informed 0%-effectiveness vs. initial self-rated effectiveness to informed 314 100%-effectiveness) as independent within-subjects variables. The results showed that 315 decreasing internal BOP (i.e., for 100% effective target faces) resulted in lower 316 subjective evaluation of others' pain and one's own unpleasantness and less monetary 317 donations in the 2 nd _ vs. 1 st _round tests, whereas enhancing BOP (i.e., for 0% effective 318 target faces) produced opposite effects ( Fig. 2d-f, see Supplementary Table 2_1 for 319 statistical details). These results replicated those in Experiment 1 and provided further 320 evidence that changing BOP resulted in variations of empathy and altruistic behavior. 321 Pearson correlations analyses of the mean rating scores in the 1 st _round and 322 2 nd _round tests across the participants showed that, for '100%-effective' patients, the Experiments 1 and 2 showed evidence that self-report measures of empathy for pain 349 were affected by BOP. In Experiment 3 we further investigated whether and how 350 changing BOP modulates brain activity in response to perceived cues signaling others' 351 pain as an objective estimation of empathy. If BOP provides a basis of empathy of 352 others' pain, decreasing BOP should reduce empathic neural responses to others' pain. 353 We tested this assumption by recording EEG to faces of 16 models from an independent 354 sample (N = 30). The participants were first presented with these faces with neutral 355 expressions and were informed that these photos were taken from 8 patients who 356 suffered from a disease and from 8 actors/actresses. The participants were asked to 357 remember patient or actor/actress identity of each neutral face and had to pass a memory 358 test with a 100% recognition accuracy. Thereafter, the participants were informed that 359 they would be presented with photos of these faces with either neutral or pain 360 expressions. Photos of pain expressions were taken from the patients who were 361 suffering from the disease whereas photos of pain expressions were taken from the 362 actors/actresses who were performing pain expressions to imitate patients' pain. The 363 participants were asked to make judgments on identity of each face (i.e., patient vs. 364 actor/actress) with a neutral or pain expression by pressing one of two buttons while 365 EEG was recorded. After EEG recording, the participants were asked to rate pain 366 intensity of each face with a pain or neutral expression on a Likert-type scale (0 = not 367 painful at all; 7 = extremely painful) and to what degree they believed in the identity of 368 each face with a pain expression on a 15-point Likert-type scale (-7 = extremely 369 believed as an actor/actress, 0 = not sure, 7 = extremely believed as a patient). Because 370 the same set of stimuli were perceived as patients or actors/actresses across the 371 participants, modulations of brain activity in response to pain expressions only reflected 372 the effects of BOP concomitant with the face identity (i.e., real pain for patients but fake 373 pain for actors/actresses). 374 The participants reported a mean positive belief score corresponding to faces with a 375 patient identity (2.496 ± 2.51) but a negative belief score corresponding to faces with 376 an actors/actresses identity (-2.210 ± 3.25) (t(29) = 4.932, p < 0.001, Cohen's d = 377 0.900, 95% CI = (2.755, 6.658)), suggesting successes of our manipulations of face 378 identities. An ANOVA of the mean rating scores of pain intensity with Identity (patient further evidence for a link between BOP and empathy for patients' pain. 389 The participants responded to face identities with high accuracies during EEG 390 recording (>81% across all conditions, see Supplementary Table 3_1 for  The experimental manipulation in Experiment 3 consisted of multiple processes, 434 including learning, memory, identity recognition, etc., to build different BOP related to 435 different faces in the participants. If BOP rather than other processes was necessary for 436 the modulation of empathic neural responses in Experiment 3, the same manipulation 437 procedure to assign different face identities that do not change BOP should change the 438 P2 amplitudes in response to pain expressions. We tested this prediction in an 439 independent sample (N = 30) in Experiment 4. We employed the stimuli and procedure 440 that were the same as those in Experiment 3 except that, during the learning phase, the 441 participants were informed that the 16 models were from two baseball teams (half from 442 a Tiger team and half from a Lion team) and they suffered from a disease. After the 443 participants had remembered team identity of each neutral face in a procedure similar to 444 that in Experiment 3, they performed identity (i.e., Tiger vs. Lion team) judgments on 445 the faces with neutral or pain expressions during EEG recording. This manipulation 446 built team identities should not influence self-report and EEG estimation of empathy 447 because the Tiger/Lion team identities did not bring any difference in BOP between 448 pain expressions of faces from the two teams. 449 The participants responded to face identities with high accuracies during EEG 450 recording (> 79% across all conditions). Rating scores of pain intensity did not differ 451 significantly between faces from the two teams (F(1,29) = 1.608, p = 0.215, η p 2 = 0.053, 452 90% CI = (0, 0.216), , Bayes factors = 0.261, Fig. 4a, see Supplementary Table 4_1 for 453 details). ANOVAs of the mean P2 amplitudes over the frontal electrodes revealed a 454 significant main effect of facial expression (F(1,29) = 12.182, P = 0.002, η p 2 = 0.296, 90% 455 CI = (0.081, 0.473), Fig. 4b and 4c, see Supplementary Table 4_2 for  times (Greenwald et al., 2003) to assess implicit associations between patient and 498 actor/actress faces and the relevant words. One-sample t-test revealed that the D score 499 was significantly larger than zero (0.929 ± 0.418, t(29) = 12.178, p < 0.001, Cohen's d = 500 2.223, 95% CI = (0.773, 1.085)), suggesting that patient faces were more strongly 501 associated with patient relevant words whereas actor/actress faces were more strongly 502 associated with actor/actress relevant words. The results indicate success of our belief 503 manipulations through the two-session tests. 504 The behavioral results in the first-session test replicated the findings of Experiment 505 1. In particular, decreasing BOP (i.e., changing patient identity in the 1 st _round test to 506 actor/actress identity in the 2 nd _round test) significantly reduced self-report of others' 507 In particular, it is unknown whether brain responses underlying cognitive and affective 573 components of empathy are similarly sensitive to the influence of BOP. Therefore, in 574 Experiment 6, we used fMRI to record BOLD signals from an independent sample (N = 575 31) to examine neural architectures in which empathic activities are modulated by BOP. 576 Similarly, the participants were first shown with photos of neutral faces of 20 models 577 and had to remember their patient (10 models) or actor/actress (10 models) identities. 578 After the participants had performed 100% correct in a memory task to recognize the 579 models' identities, they were scanned using fMRI when viewing video clips of the 580 models whose faces received painful (needle penetration) stimulation and showed pain 581  After fMRI scanning the participants were presented with each video clip again and 590 had to rate the model's pain intensity and their own unpleasantness. The participants 591 were also asked to rate the degree to which they believed in the models' patient or 592 actor/actress identities in painful video clips on a 15-point Likert-type scale (-7 = 593 extremely believed as an actor/actress, 0 = not sure, 7 = extremely believed as a patient) 594 (see Method, Supplementary Table 6_1 for  To identify neural architectures underlying BOP effects on empathy, we recorded 725 BOLD responses, using fMRI, to perceived painful and non-painful stimuli applied to 726 individuals with patient or actor/actress identities in Experiment 6. We showed that,  (Han, 2018). 794 It should be noted that our experimental manipulations changed the participants' 795 mind about the models' identities (e.g., patient vs. actor/actress) rather than explicitly 796 asking them to alter their BOP. BOP altered implicitly with target persons' identities due 797 to observers' knowledge about individuals with different identities (e.g., painful stimuli 798 applied to actors/actresses do not really hurt them and they show facial expressions to 799 pretend a specific emotional state). Therefore, the BOP effects on empathy and altruistic 800 behavior identified in our study might take place implicitly. This is different from the 801 placebo effects on first-hand pain experiences that are produced by explicitly perceived 802 verbal, conditioned, and observational cues that induce expectations of effective 803 analgesic treatments (Meissner et al., 2011). Similar explicit manipulations of making 804 individuals believe receiving oxytocin also promotes social trust and preference for 805 close social distances (Yan et al., 2018). Moreover, the placebo treatment relative to a 806 control condition significantly attenuated activations in the ACC, AI, and subcortical 807 structures (e.g., the thalamus) in response to painful electric shocks but increased the 808 prefrontal activity during anticipation of painful stimulations possibly to inhibit activity 809 in pain processing regions (Wager et al., 2004;2015). The brain regions in which 810 empathic neural responses altered due to BOP (e.g., the mPFC and TPJ) as unraveled in 811 the current study do not overlap with those in which activities are modulated by placebo 812 analgesia (Atlas and Wager, 2014). These results suggest distinct neural underpinnings 813 of BOP effects on empathic brain activity and placebo effects on brain responses to 814 first-hand pain experiences. 815 Some limitations of the current work create future research opportunities. For 816 example, a recent approach to hierarchical Bayesian models of cognition assumes that 817 the brain represents information probabilistically and people represent a state or feature 818 of the world not using a single computed value but a conditional probability density 819 function (Knill and Pouget, 2004;Friston, 2005 responses. Thirty-one Chinese students were recruited in Experiment 4 as paid 870 volunteers. One participant was excluded from data analyses due to his lower response 871 accuracy during EEG recording (< 50%). This left 30 participants (all males, 20.70 ± 872 1.97 years) for behavioral and EEG data analyses. Thirty Chinese students were 873 recruited in Experiment 5 (all males, 20.60 ± 1.75 years). Thirty-two Chinese students 874 were recruited in Experiment 6 as paid volunteers. One participant was excluded from 875 data analyses due to excessive head movement during fMRI scanning. There were 31 876 participants left (all males, 22.23 ± 2.59 years) for behavioral and fMRI data analyses. 877 The sample size in Experiment 6 was determined based on our previous fMRI research The stimuli were adopted from our previous work (Sheng and Han, 2012), which 888 consisted of photos of 16 Chinese models (half males) with each model contributing one 889 photo with pain expression and one with neutral expression. 890 After reporting demographic information, the participants were informed that they 891 would be paid with ¥10 as a basic payment for their participation. They would be able 892 to obtain an extra bonus payment as much as ¥2 depending on their decisions in the 893 following procedure. In the 1 st _round test the participants were informed that they 894 would be shown photos with pain expressions taken from patients who suffered from a 895 serious disease. After the presentation of each photo the participants were asked to 896 evaluate intensity of each patient's pain based on his/her expression by rating on a 897 Likert-type scale (0 = not painful at all; 10 = extremely painful). This rating task was 898 adopted from previous research (Jackson et al., 2005) to assess the participants' 899 understanding of others' pain feelinga key component of empathy. Immediately 900 after the pain intensity rating, the participants were asked to decide how much from the 901 extra bonus payment they would like to donate to the patient (0 to 10 points, 1 point = 902 ¥0.2). The participants were informed that the amount of one of their decisions would 903 be selected randomly and donated to one of the patients. 904 After the 1 st _round test the participants were asked to perform a short (5 mins) 905 calculation task (10 arithmetic calculations, e.g. 25-3×7=?) to clean their memory of the 906 1 st _round ratings. Thereafter, the participants were told that the photos were actually 907 taken from 8 patients and 8 actors/actresses and this experiment actually tested their 908 ability of recognizing social identities by examination of facial expressions. Faces 909 assigned with patient or actor/actress identities were counterbalanced across the 910 participants. The participants were then asked to conduct the 2 nd _round test in which 911 each photo was presented again with a word below to indicate patient or actor/actress 912 identity of the face in the photo. The participants had to report again perceived pain 913 intensity of each face and how much they would like to donate to the person shown in 914 the photo. The participants were informed that an amount of money would be finally 915 selected randomly from their 2 nd _round decisions and donated to one of the patients. 916 The total amount of the participants' donations were subject to a charity organization to 917 help patients who suffer from the same disease after the study. 918 We conducted ANOVAs of rating scores of pain intensity and amounts of monetary 919 donations with Test Phase (1 st vs. 2 nd _round) × Identity Change (patient to actor/actress 920 vs. patient to patient) as independent within-subjects variables to assess whether and 921 how beliefs of others' pain (BOP) influenced empathy and altruistic behavior towards 922 those who suffered. Finally, the participants completed two questionnaires to estimate 923 individual differences in trait empathy (Davis, 1983) and interpersonal trust (Wright and 924 Tedeschi, 1975). We analyzed the relationship between our empathy/altruistic measures 925 and individuals' trait empathy/interpersonal trust but failed to find significant results 926 and thus were not reported in the main text. 927

Mediation analysis 928
We performed mediation analyses to examine whether pain intensity mediates the 929 pathway from BOP to monetary donation. To do this, we first dummy coded decreased 930 BOP as 0 (patient identity in the 1 st _round test) and 1 (actor/actress in the 2 nd _round test) 931 or enhanced BOP as 0 (patient identity in the 1 st _round test) and 1 (patient identity was 932 confirmed in the 2 nd _round test). Then, we estimated four regression models: 1) whether 933 the independent variable (BOP) significantly accounts for the dependent variable 934 (monetary donation) when not considering the mediator (e.g., Path c); 2) whether the 935 independent variable (BOP) significantly accounts for the variance of the presumed 936 mediator (pain intensity) (e.g., Path a); 3) whether the presumed mediator (pain 937 intensity) significantly accounts for the variance of the dependent variable (monetary 938 donation) when controlling the independent variable (BOP) (e.g., Path b); 4) whether 939 the independent variable (BOP) significantly accounts for the variance of the dependent 940 variable (monetary donation) when controlling the presumed mediator (pain intensity) 941 (e.g., Path c'). To establish the mediation, the path c is not required to be significant. 942 The only requirement is that the indirect effect ab is significant. Given a significant 943 indirect effect, if Path c is not significant, the mediation is classified as indirect-only 944 mediation which is the strongest full mediation (Kenny et al., 1998;Zhao et al.., 2019). 945 A bootstrapping method was used to estimate the mediation effect. Bootstrapping is a 946 nonparametric approach to estimate effect-sizes and hypotheses of various analyses, 947 including mediation (Shrout and Bolger,2002;Mackinnon et al., 2004). Rather than 948 imposing questionable distributional assumptions, a bootstrapping analysis generates an 949 empirical approximation of the sampling distribution of a statistic by repeated random 950 resampling from the available data, which is then used to calculate p-values and 951 construct confidence intervals. 5,000 resamples were taken for our analyses. Moreover, 952 this procedure supplies superior confidence intervals (CIs) that are bias-corrected and 953 accelerated (Preacher et al., 2007;Hayes, 2008a, 2008b). The analyses 954 were performed using Hayes's PROCESS macro (Model 4, Hayes, 2017). 955

Statistical comparison 956
Behavioral data were assumed to have a normal distribution but this was not 957 formally tested. 95% Confidence intervals (95% CIs) were reported for t-tests of the 958 mean difference between two conditions and for correlation analyses of correlation 959 coefficients. 90% CIs were reported for effect sizes (η p 2 ) of ANOVA analyses. 960 According to Steiger (2004), the general rule of thumb to use CIs to test a statistical 961 hypothesis (H0) is to use a 100(1-α)% / 100(1-2α)% CI when testing a two-sided / 962 one-sided hypothesis at alpha level. We thus reported 90% CIs of η 2 in ANOVAs 963 because η 2 is always positive. 964

Experiment 2: Intrinsic BOP predicts empathy and altruistic behavior 965
The face stimuli and the procedure were the same as those in Experiment 1 except 966 the following. The participants were informed that they were to be shown photos with 967 pain expressions taken from patients who had suffered from a serious disease and 968 received medical treatment. After the presentation of each photo the participants were 969 asked to estimate how effective the medical treatment was for each patient by rating on 970 a Likert-type scale (0 = no effective or 0%, 100 = fully effective or 100% effective). 971 Besides rating pain intensity of each face in the 1 st _round test, the participants were 972 asked to report how unpleasant they were feeling when they viewed the photo (i.e., own 973 unpleasantness) by rating on a Likert-type scale (0 = not unpleasant at all, 10 = 974 extremely unpleasant). The unpleasantness rating was performed to evaluate emotional 975 sharing of others' painanother key component of empathy (Jackson et al., 2005). 976 The order of the two empathy rating tasks was counterbalanced across the participants. 977 Immediately after the empathy rating tasks, the participants were asked to decide how 978 much from the extra bonus payment they would like to donate to the patient (0 to 10 979 points, 1 point = ¥0.2). 980 In the 2 nd _round test the participants were told that the medical treatment was 981 actually effective for only half of the patients. Each photo was then presented again with 982 information that the medical treatment applied to the patient was 100% effective or 0% 983 effective. Thereafter, the participants were asked to perform the rating tasks and 984 monetary donation as those in the 1 st _round test. The participants were told that an 985 amount of money would be finally selected from their 2 nd _round decisions and donated 986 to one of the patients. 987

Mediation analysis 988
This was the same as that in Experiment 1 except that we tested whether changes of 989 pain intensity mediate the pathway from decreased BOP or enhanced BOP to changes of 990 monetary donation. To do this, we first calculated belief update (decreased BOP: 991 100%-effect minus the participants' initial estimation; enhanced BOP: the participants' 992 initial estimation minus 0%(no)-effect). Then, we estimated four regression models: 1) 993 whether the independent variable (BOP) significantly accounts for the dependent 994 variable (changes of monetary donation) when not considering the mediator (e.g., Path 995 c); 2) whether the independent variable (BOP) significantly accounts for the variance of 996 the presumed mediator (changes of pain intensity) (e.g., Path a); 3) whether the 997 presumed mediator (changes of pain intensity) significantly accounts for the variance of 998 the dependent variable (changes of monetary donation) when controlling the 999 independent variable (BOP) (e.g., Path b); 4) whether the independent variable (BOP) 1000 significantly accounts for the variance of the dependent variable (changes of monetary 1001 donation) when controlling the presumed mediator (changes of pain intensity) (e.g., 1002 Path c'). 1003

Experiment 3: Manipulated BOP changes empathic brain activity 1004
Stimuli and procedure 1005 then normalized to standard space (resampled to 3 x 3 x 3 mm 3 voxels), smoothed 1215 (FWHM= 8mm), and entered into a random effect analysis using one-sample t tests 1216 against zero. Significant results were reported using a threshold of P (FWE) < 0.05 at 1217 the voxel level. 1218

Data availability 1219
The data that support the findings of this study are available from the corresponding 1220 author upon reasonable request. 1221 Code availability 1222 The code used to analyse the data that support the findings of this study are available 1223 from the corresponding author upon reasonable request. 1224