2024.02.07.579264v1.full-html.html

Summary

Food is fundamental to survival and our brains are highly attuned to rapidly process food stimuli. Neural

signals show foods can be discriminated as edible or inedible as early as 85 ms after stimulus onset

distinguished as processed or unprocessed beginning at 130 ms

and as high or low density from 165 ms

Recent evidence revealed specialised processing of food stimuli in the ventral visual pathway

4-6

, an area that

underlies perception of faces and other important objects. For many visual objects, present perception can

be biased towards recent perceptual history (known as serial dependence

7,8

). We examined serial

dependence for food in two large samples (n>300) that rated sequences of food images for either ‘appeal’ or

‘calories’. Calorie ratings by males and females agreed closely but appeal ratings were higher in males. High

calorie ratings were associated with high appeal, especially in males. Serial analyses testing if current trial

ratings were influenced by the previous one showed both appeal and calorie ratings exhibited clear positive

dependences (i.e., a high preceding rating increased current trial ratings). The serial effect for appeal was

roughly twice that for calories and males showed a greater serial effect than females for both ratings. Serial

amplitude was larger in those who reported a longer elapsed time since they last ate and was larger in the

BMI>25 group compared to BMI<25. These findings square with recently found food selectively in visual
temporal cortex, reveal a new mechanism influencing food decision-making and suggest a new sensory-level

component that could complement cognitive strategies in diet intervention.

Results

Underlying the brain’s rapid analysis of food stimuli is a large, distributed network composed of numerous

regions, including decision-related areas in prefrontal cortex and visual areas in occipital cortex. Among

prefrontal areas, appetizing food stimuli activate several regions including orbitofrontal cortex

9-11

, medial

prefrontal cortex

9,12

and anterior cingulate

, with food saliency being linked particularly to orbitofrontal

cortex

and anticipation of pleasure from the food associated with lateral orbitofrontal cortex

. There is also

sensory-driven food-related activity in visual cortex, including the lateral occipital complex and fusiform

gyrus

3,16-18

, two visual object processing areas

. The strongest findings come from very recent studies

analysing fMRI responses to large image sets which found highly selective food specialisation in the ventral

visual pathway. Khosla et al.

found distributed activation from food images that was not due to low-level

image properties (i.e., colour, shape, texture). Jain et al.

found food-specific activity adjacent to the fusiform

face area. Another study found food images drove color-biased ventral visual areas and predicted voxel

responses

The involvement of prefrontal cortex highlights the role of choice and decision-making in food

behaviours. What to eat, when and how much are critical decisions, whether for survival (e.g., meeting

caloric needs when resources are scarce) or in affluent Western societies where a surfeit of food choices leads

us to make around 200 food decisions daily

. Following recent findings of robust food activity in ventral visual

areas

4-6

, visual perceptual factors likely also influence food choice. This is relevant as recent perceptual history

can bias current perceptual decisions so that they are not independent but biased towards recent input

(known as ‘serial dependence’

20,21

). Many visual stimuli elicit this serial bias, from basic attributes (orientation,

motion, spatial frequency, numerosity

22-25

) to more complex visual objects such as faces

26-28

, visual scenes

and even artworks

. In particular, faces – another image category with high ecological significance processed

in the fusiform gyrus – show strong serial biases for decisions about attractiveness, sex, identity and

emotion

26,27,31-34

. Given that serial dependence involves high-level (decisional) and low-level (sensory)

CC-BY-NC-ND 4.0 International license

perpetuity. It is made available under a

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 7, 2024.

;

https://doi.org/10.1101/2024.02.07.579264

doi:

bioRxiv preprint

factors

35-38

, and that food is processed in frontal decisional and visual sensory regions, food images should

also evoke serial biases. Here we examine how rating food for appeal or calorie content affects subsequent

food ratings. We find clear evidence of positive serial dependences for both calories and appeal, with current

ratings biased towards previous ratings.

Food image ratings

: Figure 2 shows ratings data for calories (2a) and appeal (2c). Data points are

mean ratings from male (blue) and female (red) participants for all 150 images. Calorie ratings show close

agreement between male and female participants and the ratings span nearly the full range of the scale.

There was no significant difference between male and female ratings for calories (males: M = 46.612;

females: M = 46.719; t

298

= .036, p = .971. See Fig 2b). Appeal ratings showed a slightly different pattern.

There was a significant difference between males and females (males: M = 54.630; females: M = 50.689; t

298

= 2.525, p = .012. See Fig 2d) and the range of ratings was reduced and clustered around the centre of the

scale (compare appeal and calories standard deviations: Fig. 2b,d). Figure 2e,f plots the mean calorie and

appeal ratings given for each of the 150 images by male (2e) and female (2f) participants. There was a

significant correlation (higher calorie ratings associated with higher appeal) in both groups, with males (r =

.559, p < .0001) showing a stronger association than females (r = .323, p = .0001).

Figure 2

. Mean ra+ngs of 150 food images for calories and appeal. Group 1 (n=330: M = 138,F = 192) made

calorie ra+ngs (2a,b) and Group 2 (n=359: M = 147,F = 212) made appeal ra+ngs (2c,d). Calorie ra+ngs did not

diﬀer between males and females and were well spread over the scale range (see 2b). Ra+ngs for food appeal

did show a signiﬁcant male/female diﬀerence (2d), with mean male appeal ra+ngs being higher. Error bars in

2b,d are ±1 SEM. Mean calorie versus appeal ra+ngs are shown in 2e,f. Each scaQer plot contains 150 data

points, being mean ra+ngs for the 150 food images by male (2e) and female (2f) par+cipants. Ra+ngs are

signiﬁcantly posi+vely correlated (higher calorie food rated as more appealing) for males and females.

Serial dependence in food ratings

: We next analysed the ratings data for serial effects to test whether

successive responses were independent or whether instead a given trial’s response was influenced by the

previous one. Figure 3 plots serial bias against relative difference. Relative difference (previous rating minus

current rating) simply quantifies whether the previous rating was less than or greater than the current rating.

Serial bias is effectively a deviation score quantifying how different the current image’s rating is from the

mean of all ratings of that image. If all ratings are sequentially independent, then serial bias will be

approximately zero for all values of relative difference. It is frequently observed, however, that perception on

a current trial is biased

towards

the previous trial (a

positive

serial dependence). In the current experiment,

100

150

Image number

100

Mean rating

Appeal ratings for each image

Males
Females

M F

0.2

0.4

0.6

0.8

Std deviation

St dev

M F

Mean

M F

Mean

M F

0.2

0.4

0.6

0.8

Std deviation

St dev

100

150

Image number

100

Mean rating

Calorie ratings for each image

Males
Females

Calorie ratings

Appeal ratings

100

Appeal

100

Calories

Female participants

r = 0.3232
p = 0.0001

100

Appeal

100

Calories

Male participants

r = 0.5594
p < .0001

Appeal rating

tin

Appeal rating

tin

CC-BY-NC-ND 4.0 International license

perpetuity. It is made available under a

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 7, 2024.

;

https://doi.org/10.1101/2024.02.07.579264

doi:

bioRxiv preprint

this would manifest as a tendency to rate images higher when the previous rating was high, and lower when

the previous rating was low.

The upper panels of Figure 3 show the serial analysis for calorie ratings. The analysis for all

participants (3a) shows a classic serial dependence effect: current trial data was biased towards higher values

when the previous trial rating was high (upper right quadrant) and was biased towards lower values when

the previous trial rating was low (bottom left quadrant). For large relative differences the effect dissipates and

returns to baseline (typical of serial dependence effects

24,35,39,40

) and is well described by a difference-of-

Gaussian (DoG) model (here, r

= 0.990) with an amplitude of 3.918 and a bandwidth of 28.769. The gray

shaded band shows the 95% confidence interval around the serial effect, calculated from 1000 iterations of

permuting the trial order and repeating the serial analysis. This provides a null distribution because the

permutation will remove any serially dependent effects and therefore indicates the effect expected by

chance. Here it is clear our serial effect is both systematic (conforming to the expected DoG function) and far

exceeds the 95% confidence limit.

Figure 3b shows the same serial analysis of the calorie ratings after splitting the group into male and

female participants. For both subgroups, the serial effect conforms well to the DoG model (males: r

= 0.988;

females: r

= 0.989) and again the shaded bands around zero serial bias show the 95% confidence interval

based on permuting the data. The DoG parameters for males and females are compared in Figure 3c which

plots amplitude and bandwidth for calorie ratings together with 95% confidence intervals (based on 1000

bootstraps of the data and refitting the DoG function each time). Males and females did not differ in the

bandwidth parameter (the range of relative difference over which the serial effect occurs: Males = 28.307,

females = 28.987, p = .829) but the amplitude of the serial effect differed significantly, with males showing a

larger serial effect (males = 4.341, females = 3.656, p < .001).

The lower panels of Figure 3 show the serial analysis for appeal ratings. Figure 3d confirms the serial

effect also occurs for rating food on appeal and indeed the effect is larger in amplitude and broader in range

than was observed for calorie ratings (amplitude = 7.316, bandwidth = 52.123, r

= 0.993). Figure 3e shows

the serial effect for appeal ratings separately for male and female participants. The effect again conforms well

to a DoG function (male: r

= 0.997, female: r

= 0.988) and far exceeds the 95% confidence interval around

the null distribution (shaded bands in 3d,e). There is again a larger serial effect for males than for females

(males = 8.671, females = 6.343, p < .001) and also a difference in bandwidth, with males showing a serial

effect over a broader range (males = 55.306, females = 48.664, p = .016). Figure 3f summarises the amplitude

and bandwidth parameters from the best-fitting DoG functions to male and female appeal ratings with 95%

confidence intervals.

CC-BY-NC-ND 4.0 International license

perpetuity. It is made available under a

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 7, 2024.

;

https://doi.org/10.1101/2024.02.07.579264

doi:

bioRxiv preprint

Figure 3

. Food ra+ngs for calories and appeal show a posi+ve serial dependence. a) Calorie ra+ngs deviate

posi+vely from their mean value when the previous food was rated highly (and nega+vely when the previous

food was rated low). This aQrac+on to the previous food’s ra+ng returns toward baseline for large diﬀerences

between current and previous trial ra+ngs and is well described by a Diﬀerence of Gaussian (DoG) model. The

shaded band around zero bias shows the null distribu+on for serial bias produced by permu+ng the data 1000

+mes and reﬁWng a DoG each +me. Permu+ng trial order removes any sequen+al eﬀects and the band shows

the central 95% of the null distribu+on. b) Group calorie ra+ngs from 3a split into male and female subgroups

ﬁQed with DoG func+ons. c) DoG parameters compared for male and female groups. Columns show the mean

amplitude and bandwidth based on 1000 bootstraps of the data and a reﬁWng of the DoG on each itera+on.

d,e,f) Appeal data: Lower panels contain the same analyses as the upper panels. The summary in 3f shows the

serial eﬀect is larger overall for appeal than for calories and is stronger for males in both amplitude and

bandwidth. Error bars in all panels show 95% conﬁdence intervals.

Finally, we analysed demographic and self-report ratings provided prior to the experiment.

Participants gave their age, height and weight (from which BMI was calculated), and whether they were

currently dieting (yes/no), as well as answering the following questions on a 0-100 scale: How

hungry

are you

now? How

thirsty

are you now? How

tired

are you now? They also indicated in hours: How long since your

last meal? How long since you last ate? To test if any of these measures modulated the serial effect, we

divided the sample into two subgroups using a median split (e.g., high hunger vs low hunger). The exceptions

were dieting, which was split based on a binary yes/no response, and BMI where we compared BMI > 25
with BMI <25. We conducted serial analyses separately on the high and low subgroups for each variable and

determined if the strength of the serial effect differed by subtracting the amplitude of the low group from the

high group. Results are shown in Figure 4 and any value greater than zero indicates a stronger serial effect in

the ‘high’ subgroup.

-100

-50

100

Relative difference (previous minus current)

-10

-5

Serial bias

Serial response effect. DV = Calories.

-100

-50

100

Relative difference (previous minus current)

-10

-5

Serial bias

Serial response effect. DV = Appeal.

-100

-50

100

Relative difference (previous minus current)

-10

-5

Serial bias

Serial response effect. DV = Appeal.

-100

-50

100

Relative difference (previous minus current)

-10

-5

Serial bias

Serial response effect. DV = Calories.

M F

Serial amplitude

Serial bandwidth

M F

Serial amplitude

Serial bandwidth

DoG parameters

Calories

DoG parameters

Appeal

Serial effect: Calories (all participants)

Serial effect: Calories (male vs female)

Serial effect: Appeal (all participants)

Serial effect: Appeal (male vs female)

CC-BY-NC-ND 4.0 International license

perpetuity. It is made available under a

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 7, 2024.

;

https://doi.org/10.1101/2024.02.07.579264

doi:

bioRxiv preprint

Figure 4

. Results of dividing the sample using a median split. For each of the eight variables shown, the sample

was divided into two subgroups and serial analyses were conducted for each group. Columns show the

diﬀerence in the amplitude of the serial eﬀect (high group minus low group), with values greater than zero

indica+ng a stronger serial eﬀect in the high group. A) For appeal ra+ngs, there were strong and robust

increases in serial amplitude for the high BMI group (BMI >25 vs BMI < 25) and in the groups that reported

longer periods since they last ate or had a meal. B) Calorie ra+ngs were much more stable between high and

low groups for all eight variables, although the high sta+s+cal power of our large sample meant some small

diﬀerences were signiﬁcant. All error bars show 95% conﬁdence intervals based on 1000 bootstraps.

Discussion

This study was motivated by very recent findings showing that visual areas in human temporal cortex show a
specialisation for food images

4-6

. We examined whether participants’ ratings of food images would exhibit a

bias known as serial dependence, just as occurs with ratings of faces, another prominent visual specialisation

in temporal cortex

41,42

and for many other visual attributes

. Ratings of food images sequences for calories

and appeal both showed significant positive serial dependences: rather than being sequentially independent,

ratings tended to follow the rating given to the preceding image. A highly appealing preceding food image led

to a current food image being rated higher than its mean appeal rating (and

vice versa

: a preceding low-

appeal image made a current image less appealing). Rating food images for calories showed the same

pattern of positive serial bias, although appeal ratings showed a stronger serial effect.

Serial dependence has been recently an active field in perceptual decision making. This attractive

bias is an adaptive change in sensory information that makes current perception assimilate towards

previously seen stimuli. It is widely reported in perceptual judgements about many stimuli from simple (e.g.,

such as orientation and spatial frequency

23,24

) to complex judgements about face perception (e.g., gender,

attractiveness, identity

), scene perception

, and aesthetics

. Serial dependence also occurs at the level of

global information where elements are combined into ensemble objects

and may occur separately for

individual objects in a multi-object scene

37,43

. It is also a real world-effect: when realistic movie clips are used,

current perception of objects is biased by information presented up to 15 seconds earlier

and current

emotion is biased for up to 12 seconds

. These findings suggest that food serial dependence is likely to occur

in real-world contexts containing multiple food options and future work could test this.

CC-BY-NC-ND 4.0 International license

perpetuity. It is made available under a

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 7, 2024.

;

https://doi.org/10.1101/2024.02.07.579264

doi:

bioRxiv preprint

Food appeal showed a larger serial effect than food calorie ratings. Attention plays a role in boosting

serial dependence effects

7,21

and appealing foods capture attention, likely to a greater extent than an

analytical food assessment such as rating calories. A related question is why serial dependence is greater in

males. Perhaps males are more attentionally biased to food, driving a stronger serial effect. A recent review of

the large literature on food attention bias notes few studies have reliable data on gender differences

and

some find no gender differences

. A systematic review found that food attentional bias is most prevalent in

obese individuals

, which implies our gender differences are not BMI-related as median BMI scores were

similar for males (24.8) and females (23.0). Follow-up work is needed to examine why food serial

dependence differs depending on the rater’s gender and the food variable being rated.

Food decision-making is modulated by many factors

49-51

. We therefore collected data from

participants on variables such as age, BMI, hunger, tiredness, etc. Comparing serial effects after a median split

on each variable, the largest differences between high and low groups occurred for food appeal ratings (Fig.

4a), with the high BMI (>25) group showing greater serial dependence (nearly a factor of 2) than the low

group and greater time since last eating or consuming a meal also driving stronger serial dependence. These

findings fit with BMI exerting a direct influence on the brain’s food network

52,53

and with elapsed time

increasing hunger and thus food network activity

and ghrelin levels (a hormone triggering hunger sensation

and food seeking behaviours that increases neural response in orbitofrontal and visual cortex cortices

Given the high frequency of food decisions in our daily lives, serial dependence analyses are perfectly

suited to uncovering drivers of food decision-making. We are surrounded by food options in affluent western

countries, many highly processed with unhealthy combinations of calories, fat, sugar and salt. High ultra-

processed food intake is linked with poor physical health (cardiovascular disease, cancer and overall mortality

rate

56-58

) and mental health outcomes, with a recent meta-analysis finding greater odds of depressive and

anxiety symptoms and increased risk of subsequent depression

. There is also evidence that such foods can

induce eating addiction

60-62

. Understanding all aspects of food decision-making is therefore critical and our

novel finding that food decisions exhibit positive serial dependence reveals a new component, one that adds

a positive feedback loop making food look more appealing following a preceding appealing food, especially in

overweight/obese individuals or those who have not eaten for several hours, and thus driving consumption.

Conversely, food serial dependence makes a food less appealing when it follows a food rated low in appeal.

Equivalently, the calorie rating given to a food could be boosted or attenuated by the preceding food. These

findings could help guide clinical interventions, adding a sensory-based component to predominantly

cognitive-based strategies designed to either reduce food intake (e.g., when treating obesity or compulsion to

eat) or increase it (e.g., when treating bulimia or anorexia nervosa).

CC-BY-NC-ND 4.0 International license

perpetuity. It is made available under a

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 7, 2024.

;

https://doi.org/10.1101/2024.02.07.579264

doi:

bioRxiv preprint

Methods

Participants

Experiment 1 measured calorie ratings and involved 330 participants; 138 males (42% of

sample) and 192 females (58%). Experiment 2 measured appeal ratings and involved 359 participants; 147

males (41% of sample) and 212 females (59%). Recruitment was made online through Prolific and there

were no exclusions during recruitment or data analysis.

Figure 1

. S+muli and methods. (A) 150 food images were selected from the ‘transformed’ and ‘natural food’

categories of the FRIDa image set (

hQps://foodcast.sissa.it/neuroscience

). The set of 150 images was presented

in a random order and then presented again two more +mes in new random orders (3 ra+ngs per image). (B)

Each trial began with a blank screen averaging 750 ms (±200 ms random jiQer), then a food image (250 ms)

then a blank screen (1000 ms ±200), then a response slider that par+cipants controlled with keyboard arrow

keys to indicate their ra+ng of the food (recorded with a mouse click). (C) Two groups were recruited from

Proliﬁc so that two dependent variables could be measured. Group 1 (n = 330) rated the calorie value of the

food; group 2 (n = 359) rated the appeal of the food.

Stimuli

Food images were selected from the FRIDa image set (

https://foodcast.sissa.it/neuroscience

). We

used 150 images drawn from the ‘transformed’ and ‘natural food’ categories (75 per category) that were
selected to approximately evenly span the calorie range. These are all color images, cropped and presented

on a white background, and were standardised to a size of 375 x 375 pixels and a resolution of 72 dpi.

Stimulus presentation duration 250 ms.

Procedure

The procedure was the same for Experiments 1 and 2. After instructions and practice trials to

become familiar with the procedure participants were presented with 450 trials (a set of 150 food images

presented three times with each set in a new random order). The trial sequence was: (i) a blank screen with a

fixation cross for 750 ms (randomly varied in the range of ±200 ms), (ii) an image presentation for 250 ms, (iii)

a blank fixation screen again for 1000 ms (randomly varied within ±200 ms), and (iv) a keyboard-driven ratings

bar used to record the participant’s response. The ratings bar ranged from 0 – 100 with the slider initially set

to 50 and needing to be adjusted before a response was accepted. In Experiment 1 (n=330) the task was to

CC-BY-NC-ND 4.0 International license

perpetuity. It is made available under a

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 7, 2024.

;

https://doi.org/10.1101/2024.02.07.579264

doi:

bioRxiv preprint

rate the foods for calories. In Experiment 2 (a different sample of n=359) the task was to rate the foods for

appeal.

Demographic data were collected for every participant (age, gender, height, weight) and responses

to the following: rate your hunger; rate your thirst; rate your tiredness; are you currently dieting?; how long

since you last ate (hours)?; how long since your last meal (hours)?

Design and data analysis

The aim of the study was to measure ratings for a large number of food images

(n=150). Collecting data online meant that a large number of participants could be reached so that variability

in demographic variables such as gender and BMI could be analysed. The use of a large set of food images

has the potential limitation of relatively few ratings per image (3 ratings each) yet having a large sample

means this is easily overcome by analysing data as single aggregate subject (a ‘super subject’ analysis) and

using a bootstrapping procedure to obtain measures of variance and to calculate confidence intervals.

CC-BY-NC-ND 4.0 International license

perpetuity. It is made available under a

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 7, 2024.

;

https://doi.org/10.1101/2024.02.07.579264

doi:

bioRxiv preprint

References

Tsourides, K.

et al.

Neural correlates of the food/non-food visual dislnclon.

Biol

Psychol

115

, 35-42, doi:10.1016/j.biopsycho.2015.12.013 (2016).

Coricelli, C., Toepel, U., Nomer, M. L., Murray, M. M. & Rumial, R. I. Dislnct brain

representalons of processed and unprocessed foods.

Eur J Neurosci

, 3389-3401,

doi:10.1111/ejn.14498 (2019).

Toepel, U., Knebel, J. F., Hudry, J., le Coutre, J. & Murray, M. M. The brain tracks the

energelc value in food images.

Neuroimage

, 967-974,

doi:10.1016/j.neuroimage.2008.10.005 (2009).

Jain, N.

et al.

Seleclvity for food in human ventral visual cortex.

Commun Biol

, 175,

doi:10.1038/s42003-023-04546-2 (2023).

Khosla, M., Ratan Murty, N. A. & Kanwisher, N. A highly seleclve response to food in

human visual cortex revealed by hypothesis-free voxel decomposilon.

Curr Biol

4159-4171 e4159, doi:10.1016/j.cub.2022.08.009 (2022).

Pennock, I. M. L.

et al.

Color-biased regions in the ventral visual pathway are food

seleclve.

Curr Biol

, 134-146 e134, doi:10.1016/j.cub.2022.11.063 (2023).

Cicchini, G. M., Mikellidou, K. & Burr, D. C. Serial Dependence in Perceplon.

Annu

Rev Psychol

, doi:10.1146/annurev-psych-021523-104939 (2023).

Manassi, M., Murai, Y. & Whitney, D. Serial dependence in visual perceplon: A meta-

analysis and review.

J Vis

, 18, doi:10.1167/jov.23.8.18 (2023).

Killgore, W. D. & Yurgelun-Todd, D. A. Body mass predicts orbitofrontal aclvity during

visual presentalons of high-calorie foods.

Neuroreport

, 859-863,

doi:10.1097/00001756-200505310-00016 (2005).

Mengoq, P., Foroni, F. & Rumial, R. I. Neural correlates of the energelc value of

food during visual processing and response inhibilon.

Neuroimage

184

, 130-139,

doi:10.1016/j.neuroimage.2018.09.017 (2019).

Simmons, W. K., Marln, A. & Barsalou, L. W. Pictures of appelzing foods aclvate

gustatory corlces for taste and reward.

Cereb Cortex

, 1602-1608,

doi:10.1093/cercor/bhi038 (2005).

Goldstone, A. P.

et al.

Faslng biases brain reward systems towards high-calorie foods.

Eur J Neurosci

, 1625-1635, doi:10.1111/j.1460-9568.2009.06949.x (2009).

Frank, S.

et al.

Processing of food pictures: inﬂuence of hunger, gender and calorie

content.

Brain Res

1350

, 159-166, doi:10.1016/j.brainres.2010.04.030 (2010).

Plassmann, H., O'Doherty, J. P. & Rangel, A. Appellve and aversive goal values are

encoded in the medial orbitofrontal cortex at the lme of decision making.

J Neurosci

, 10799-10808, doi:10.1523/JNEUROSCI.0788-10.2010 (2010).

van der Laan, L. N., de Ridder, D. T., Viergever, M. A. & Smeets, P. A. The ﬁrst taste is

always with the eyes: a meta-analysis on the neural correlates of processing visual

food cues.

Neuroimage

, 296-303, doi:10.1016/j.neuroimage.2010.11.055 (2011).

Adamson, K. & Troiani, V. Dislnct and overlapping fusiform aclvalon to faces and

food.

Neuroimage

174

, 393-406, doi:10.1016/j.neuroimage.2018.02.064 (2018).

Grootswagers, T., Cichy, R. M. & Carlson, T. A. Finding decodable informalon that can

be read out in behaviour.

Neuroimage

179

, 252-262,

doi:10.1016/j.neuroimage.2018.06.022 (2018).

Huerta, C. I., Sarkar, P. R., Duong, T. Q., Laird, A. R. & Fox, P. T. Neural bases of food

perceplon: coordinate-based meta-analyses of neuroimaging studies in mullple

modaliles.

Obesity (Silver Spring)

, 1439-1446, doi:10.1002/oby.20659 (2014).

CC-BY-NC-ND 4.0 International license

perpetuity. It is made available under a

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 7, 2024.

;

https://doi.org/10.1101/2024.02.07.579264

doi:

bioRxiv preprint

Wansink, B. & Sobal, J. Mindless ealng: The 200 daily food decisions we overlook.

Environment and Behavior

, 106-123, doi:10.1177/0013916506295573 (2007).

Kiyonaga, A., Scimeca, J. M., Bliss, D. P. & Whitney, D. Serial Dependence across

Perceplon, Amenlon, and Memory.

Trends Cogn Sci

, 493-497,

doi:10.1016/j.lcs.2017.04.011 (2017).

Pascucci, D.

et al.

Serial dependence in visual perceplon: A review.

J Vis

, 9,

doi:10.1167/jov.23.1.9 (2023).

Alais, D., Leung, J. & Van der Burg, E. Linear Summalon of Repulsive and Amraclve

Serial Dependencies: Orientalon and Molon Dependencies Sum in Molon

Perceplon.

J Neurosci

, 4381-4390, doi:10.1523/JNEUROSCI.4601-15.2017 (2017).

Cicchini, G. M., Mikellidou, K. & Burr, D. C. The funclonal role of serial dependence.

Proc Biol Sci

285

, doi:10.1098/rspb.2018.1722 (2018).

Fischer, J. & Whitney, D. Serial dependence in visual perceplon.

Nature neuroscience

, 738-743, doi:10.1038/nn.3689 (2014).

Fornaciai, M. & Park, J. Serial dependence in numerosity perceplon.

J Vis

, 15,

doi:10.1167/18.9.15 (2018).

Liberman, A., Fischer, J. & Whitney, D. Serial dependence in the perceplon of faces.

Curr Biol

, 2569-2574, doi:10.1016/j.cub.2014.09.025 (2014).

Taubert, J., Alais, D. & Burr, D. Diﬀerent coding strategies for the perceplon of stable

and changeable facial amributes.

ScienFﬁc reports

, 32239, doi:10.1038/srep32239

(2016).

Van der Burg, E., Rhodes, G. & Alais, D. Posilve sequenlal dependency for face

amraclveness perceplon.

J Vis

, 6, doi:10.1167/19.12.6 (2019).

Manassi, M., Liberman, A., Chaney, W. & Whitney, D. The perceived stability of

scenes: serial dependence in ensemble representalons.

ScienFﬁc reports

, 1971,

doi:10.1038/s41598-017-02201-5 (2017).

Kim, S., Burr, D. & Alais, D. Amraclon to the recent past in aesthelc judgments: A

posilve serial dependence for ralng artwork.

J Vis

, 19, doi:10.1167/19.12.19

(2019).

Kok, R., Taubert, J., Van der Burg, E., Rhodes, G. & Alais, D. Face familiarity promotes

stable idenlty recognilon: exploring face perceplon using serial dependence.

R Soc

Open Sci

, 160685, doi:10.1098/rsos.160685 (2017).

Liberman, A., Manassi, M. & Whitney, D. Serial dependence promotes the stability of

perceived emolonal expression depending on face similarity.

AHen Percept

Psychophys

, 1461-1473, doi:10.3758/s13414-018-1533-8 (2018).

Taubert, J., Van der Burg, E. & Alais, D. Love at second sight: Sequenlal dependence

of facial amraclveness in an on-line dalng paradigm.

ScienFﬁc reports

, 22740,

doi:10.1038/srep22740 (2016).

Turbem, K., Palermo, R., Bell, J., Burton, J. & Jeﬀery, L. Individual Diﬀerences in Serial

Dependence of Facial Idenlty are Associated with Face Recognilon Abililes.

ScienFﬁc reports

, 18020, doi:10.1038/s41598-019-53282-3 (2019).

Ceylan, G., Herzog, M. H. & Pascucci, D. Serial dependence does not originate from

low-level visual processing.

CogniFon

212

, 104709,

doi:10.1016/j.cognilon.2021.104709 (2021).

Cicchini, G. M., Mikellidou, K. & Burr, D. Serial dependencies act directly on

perceplon.

J Vis

, 6, doi:10.1167/17.14.6 (2017).

CC-BY-NC-ND 4.0 International license

perpetuity. It is made available under a

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 7, 2024.

;

https://doi.org/10.1101/2024.02.07.579264

doi:

bioRxiv preprint

Collins, T. Serial dependence tracks objects and scenes in parallel and independently.

J Vis

, 4, doi:10.1167/jov.22.7.4 (2022).

St John-Saallnk, E., Kok, P., Lau, H. C. & de Lange, F. P. Serial Dependence in

Perceptual Decisions Is Reﬂected in Aclvity Pamerns in Primary Visual Cortex.

Neurosci

, 6186-6192, doi:10.1523/JNEUROSCI.4390-15.2016 (2016).

Alais, D., Xu, Y., Wardle, S. G. & Taubert, J. A shared mechanism for facial expression

in human faces and face pareidolia.

Proc Biol Sci

288

, 20210966,

doi:10.1098/rspb.2021.0966 (2021).

Fritsche, M. & de Lange, F. P. The role of feature-based amenlon in visual serial

dependence.

J Vis

, 21, doi:10.1167/19.13.21 (2019).

Kanwisher, N. & Yovel, G. The fusiform face area: a corlcal region specialized for the

perceplon of faces.

Philos Trans R Soc Lond B Biol Sci

361

, 2109-2128,

doi:10.1098/rstb.2006.1934 (2006).

Tsao, D. The Macaque Face Patch System: A Window into Object Representalon.

Cold

Spring Harb Symp Quant Biol

, 109-114, doi:10.1101/sqb.2014.79.024950 (2014).

Fischer, C.

et al.

Context informalon supports serial dependence of mullple visual

objects across memory episodes.

Nat Commun

, 1932, doi:10.1038/s41467-020-

15874-w (2020).

Manassi, M. & Whitney, D. Illusion of visual stability through aclve perceptual serial

dependence.

Sci Adv

, eabk2480, doi:10.1126/sciadv.abk2480 (2022).

Ortega, J., Chen, Z. & Whitney, D. Serial dependence in emolon perceplon mirrors

the autocorrelalons in natural emolon stalslcs.

J Vis

, 12,

doi:10.1167/jov.23.3.12 (2023).

Stom, N., Fox, J. R. E. & Williams, M. O. Amenlonal bias in ealng disorders: A meta-

review.

Int J Eat Disord

, 1377-1399, doi:10.1002/eat.23560 (2021).

Hardman, C. A., Rogers, P. J., Etchells, K. A., Houstoun, K. V. & Munafo, M. R. The

eﬀects of food-related amenlonal bias training on appelte and food intake.

AppeFte

, 295-300, doi:10.1016/j.appet.2013.08.021 (2013).

Hendrikse, J. J.

et al.

Amenlonal biases for food cues in overweight and individuals

with obesity: a systemalc review of the literature.

Obes Rev

, 424-432,

doi:10.1111/obr.12265 (2015).

Leng, G.

et al.

The determinants of food choice.

Proc Nutr Soc

, 316-327,

doi:10.1017/S002966511600286X (2017).

Sobal, J., Bisogni, C. A. & Jastran, M. Food Choice Is Mullfaceted, Contextual,

Dynamic, Mulllevel, Integrated, and Diverse.

Mind, Brain and EducaFon

, 6-12

(2014).

Steptoe, A., Pollard, T. M. & Wardle, J. Development of a measure of the molves

underlying the seleclon of food: the food choice queslonnaire.

AppeFte

, 267-

284, doi:10.1006/appe.1995.0061 (1995).

Garcia-Garcia, I.

et al.

Neural responses to visual food cues: insights from funclonal

magnelc resonance imaging.

Eur Eat Disord Rev

, 89-98, doi:10.1002/erv.2216

(2013).

Kennedy, J. & Dimitropoulos, A. Inﬂuence of feeding state on neurofunclonal
diﬀerences between individuals who are obese and normal weight: a meta-analysis

of neuroimaging studies.

AppeFte

, 103-109, doi:10.1016/j.appet.2013.12.017

(2014).

CC-BY-NC-ND 4.0 International license

perpetuity. It is made available under a

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 7, 2024.

;

https://doi.org/10.1101/2024.02.07.579264

doi:

bioRxiv preprint

Dicker-Oren, S. D., Gelkopf, M. & Greene, T. The dynamic network associalons of

food craving, restrained ealng, hunger and negalve emolons.

AppeFte

175

106019, doi:10.1016/j.appet.2022.106019 (2022).

Malik, S., McGlone, F., Bedrossian, D. & Dagher, A. Ghrelin modulates brain aclvity in

areas that control appellve behavior.

Cell Metab

, 400-409,

doi:10.1016/j.cmet.2008.03.007 (2008).

Isaksen, I. M. & Dankel, S. N. Ultra-processed food consumplon and cancer risk: A

systemalc review and meta-analysis.

Clin Nutr

, 919-928,

doi:10.1016/j.clnu.2023.03.018 (2023).

Kim, H., Hu, E. A. & Rebholz, C. M. Ultra-processed food intake and mortality in the

USA: results from the Third Nalonal Health and Nutrilon Examinalon Survey

(NHANES III, 1988-1994).

Public Health Nutr

, 1777-1785,

doi:10.1017/S1368980018003890 (2019).

Srour, B.

et al.

Ultra-processed food intake and risk of cardiovascular disease:

prospeclve cohort study (NutriNet-Sante).

BMJ

365

, l1451, doi:10.1136/bmj.l1451

(2019).

Lane, M. M.

et al.

Ultra-Processed Food Consumplon and Mental Health: A

Systemalc Review and Meta-Analysis of Observalonal Studies.

Nutrients

doi:10.3390/nu14132568 (2022).

Hebebrand, J.

et al.

"Ealng addiclon", rather than "food addiclon", bemer captures

addiclve-like ealng behavior.

Neurosci Biobehav Rev

, 295-306,

doi:10.1016/j.neubiorev.2014.08.016 (2014).

Richmond, R. L., Roberto, C. A. & Gearhardt, A. N. The associalon of addiclve-like

ealng with food intake in children.

AppeFte

117

, 82-90,

doi:10.1016/j.appet.2017.06.002 (2017).

Schulte, E. M., Avena, N. M. & Gearhardt, A. N. Which foods may be addiclve? The

roles of processing, fat content, and glycemic load.

PLoS One

, e0117959,

doi:10.1371/journal.pone.0117959 (2015).

CC-BY-NC-ND 4.0 International license

perpetuity. It is made available under a

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 7, 2024.

;

https://doi.org/10.1101/2024.02.07.579264

doi:

bioRxiv preprint

STAR

★

Methods

Key resources table

REAGENT or RESOURCE

SOURCE

IDENTIFIER

Deposited data

Psychophysical data

This paper

To be deposit4d on OSF (URL)

SoMware and algorithms

Online code (TC to upload)

MATLAB R2021a

Psychtoolbox-3

This paper

To be deposited on OSF (URL)

RRID:

SCR_001622

RRID:

SCR_002881

Resource availability

Lead contact

Further informalon and requests for resources should be directed to and will be fulﬁlled by

the lead contact, David Alais (

david.alais@sydney.edu.au

Materials availability

This study did not generate new unique reagents

Data and code availability

Data reported in this study and the custom code for analyses are deposited at Open Science

Framework (OSF URL). Custom code for experiment and visualizalon are available by

request to the lead contact.

Experiment model and parScipant details

Experiment 1 measured calorie ralngs and involved 330 parlcipants; 138 males (42% of

sample) and 192 females (58%). Experiment 2 measured appeal ralngs and involved 359

parlcipants; 147 males (41% of sample) and 212 females (59%). Recruitment was made

online through Proliﬁc and there were no exclusions during recruitment or data analysis.

Method details

General methods

SSmuli

Food images were selected from the FRIDa image set

(

hmps://foodcast.sissa.it/neuroscience

). We used 150 images drawn from the ‘transformed’

and ‘natural food’ categories (75 per category) that were selected to approximately evenly

span the calorie range. These are all color images, cropped and presented on a white

background, and were standardised to a size of 375 x 375 pixels and a resolulon of 72 dpi.

Slmulus presentalon duralon 250 ms.

Procedure

The procedure was the same for Experiments 1 and 2. A}er instruclons and

praclce trials to become familiar with the procedure parlcipants were presented with 450

trials (a set of 150 food images presented three lmes with each set in a new random order).

The trial sequence was: (i) a blank screen with a ﬁxalon cross for 750 ms (randomly varied in

the range of ±200 ms), (ii) an image presentalon for 250 ms, (iii) a blank ﬁxalon screen

CC-BY-NC-ND 4.0 International license

perpetuity. It is made available under a

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 7, 2024.

;

https://doi.org/10.1101/2024.02.07.579264

doi:

bioRxiv preprint

again for 1000 ms (randomly varied within ±200 ms), and (iv) a keyboard-driven ralngs bar

used to record the parlcipant’s response. The ralngs bar ranged from 0 – 100 with the

slider inilally set to 50 and needing to be adjusted before a response was accepted. In

Experiment 1 (n=330) the task was to rate the foods for calories. In Experiment 2 (a diﬀerent

sample of n=359) the task was to rate the foods for appeal.

Demographic data were collected for every parlcipant (age, gender, height, weight)

and responses to the following: rate your hunger; rate your thirst; rate your lredness; are

you currently dielng?; how long since you last ate (hours)?; how long since your last meal

(hours)?

Design and data analysis

The aim of the study was to measure ralngs for a large number of

food images (n=150). Colleclng data online meant that a large number of parlcipants could

be reached so that variability in demographic variables such as gender and BMI could be

analysed. The use of a large set of food images has the potenlal limitalon of relalvely few

ralngs per image (3 ralngs each) yet having a large sample means this is easily overcome by
analysing data as single aggregate subject (a ‘super subject’ analysis) and using a

bootstrapping procedure to obtain measures of variance and to calculate conﬁdence

intervals.

Acknowledgments

This research was supported by a grant from the Australian Research Council (DP210101691)

awarded to D.A..

Author contribuSons

D.A. conceived of the study and D.A. and T.C. designed the research. T.C. implemented the

experiment for online data colleclon. D.A. analysed and interpreted the data, prepared the

ﬁgures and wrote the paper with feedback on the manuscript dra} from T.C..

DeclaraSon of interests

The authors declare no compelng interests.

Supplemental informaSon

None.

CC-BY-NC-ND 4.0 International license

perpetuity. It is made available under a

preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in

The copyright holder for this

this version posted February 7, 2024.

;

https://doi.org/10.1101/2024.02.07.579264

doi:

bioRxiv preprint