Resource

https://doi.org/10.1038/s41593-019-0436-x

Functional boundaries in the human cerebellum

revealed by a multi-domain task battery
Maedbh King1,2, Carlos R. Hernandez-Castillo2, Russell A. Poldrack3, Richard B. Ivry1 and
Jörn Diedrichsen 2,4,5*

There is compelling evidence that the human cerebellum is engaged in a wide array of motor and cognitive tasks. A fundamen-
tal question centers on whether the cerebellum is organized into distinct functional subregions. To address this question, we
employed a rich task battery designed to tap into a broad range of cognitive processes. During four functional MRI sessions,
participants performed a battery of 26 diverse tasks comprising 47 unique conditions. Using the data from this multi-domain
task battery, we derived a comprehensive functional parcellation of the cerebellar cortex and evaluated it by predicting func-
tional boundaries in a novel set of tasks. The new parcellation successfully identified distinct functional subregions, providing
significant improvements over existing parcellations derived from task-free data. Lobular boundaries, commonly used to sum-
marize functional data, did not coincide with functional subdivisions. The new parcellation provides a functional atlas to guide
future neuroimaging studies.

C
onverging lines of research provide compelling evidence that engage a broad range of sensorimotor, cognitive and social/affective
the cerebellum is engaged in a broad range of cognitive func- processes. Using a block design, activation for each task was mea-
tions, well beyond its historical association with sensorimotor sured over four fMRI scanning sessions against a common baseline.
control1. Anatomical tracing studies in non-human primates have Our task set was successful in eliciting activation across the entirety
revealed reciprocal connections with parietal and prefrontal asso- of the cerebellar cortex, allowing us to derive a novel parcellation
ciation cortices2. Individuals with lesions to the cerebellum exhibit that characterizes the functional profile of cerebellar subregions in
behavioral impairments on tasks designed to assess non-motor pro- unprecedented detail. The breadth of the task sets also enabled us to
cesses such as duration discrimination, attentional control, spatial summarize the functional specialization of each region in terms of
cognition, emotion perception and executive and language func- the underlying latent motor, cognitive and social/affective features.
tion. Perhaps most intriguing, neuroimaging studies consistently We developed a metric to evaluate the strength of the proposed
reveal activations of the cerebellar cortex during a diverse set of functional boundaries. This allowed us to address the fundamen-
motor, cognitive, and social and affective tasks3. tal question of whether there are distinct functional regions in the
This raises the question of whether the cerebellum can be mean- cerebellum, or whether functional specialization is better described
ingfully subdivided into a discrete set of regions, reflecting distinct in terms of continuous gradients7. The approach is predicated on
functional contributions across diverse task domains. In contrast to the idea that, if a boundary between two regions divides function-
the cerebral cortex, the cytoarchitectonic organization is remark- ally heterogeneous regions, then the activation pattern for two vox-
ably uniform across the entire cerebellar cortex. Due to this homo- els that lie within the same region should be more correlated than
geneity, neuroimaging and neuropsychological studies have mostly voxel pairs that span a boundary. Critically, a meaningful functional
relied on the macroanatomical folding of the cerebellum along the parcellation needs to be predictive of boundaries for the activation
superior to inferior axis into ten lobules (numbered I–X)4. More patterns elicited by a different set of tasks. Using this approach, we
recently, functional parcellations based on task-free functional demonstrate that the cerebellum has discrete functional regions and
magnetic resonance imaging (fMRI) data have been proposed5–7. that our multi-domain task battery (MDTB) parcellation is superior
However, the degree to which these proposed boundaries corre- to alternatives in predicting functional boundaries. The new func-
spond to functional divisions remains unclear. Task-based studies tional parcellation of the cerebellar cortex provides an important
have been limited by the lack of a comprehensive neuroimaging step toward understanding the role of the cerebellum across diverse
data set. A few studies have employed data sets involving multiple functional domains.
tasks7,8, but the small number of task conditions (<7) and the lack
of a common measurement baseline have made it difficult to derive Results
and evaluate task-based functional parcellations. The functional To obtain a comprehensive functional parcellation of the cerebellar
heterogeneity of the cerebellum has also been explored using meta- cortex, we developed an MDTB of 26 tasks comprising 47 unique
analytic approaches9, which have the disadvantage that data for dif- task conditions (Fig. 1a and Supplementary Table 1), selected to
ferent tasks come from different groups of participants. encompass a wide range of processes required for motor, cogni-
In the present study, we aimed to fully characterize the func- tive and affective/social function. To avoid strong learning-related
tional organization of the cerebellar cortex by employing a large and changes, 24 healthy individuals were trained on the task protocol
diverse task battery comprising 47 unique conditions designed to (approximately 10 h) before scanning. During scanning, each task

Department of Psychology, University of California, CA, Berkeley, USA. 2Brain and Mind Institute, Western University, Ontario, London, Canada.
1

Department of Psychology, Stanford University, CA, Stanford, USA. 4Department of Statistical and Actuarial Sciences, Western University, London,
3

Ontario, Canada. 5Department of Computer Science, Western University, London, Ontario, Canada. *e-mail: [email protected]

Nature Neuroscience | VOL 22 | AUGUST 2019 | 1371–1378 | www.nature.com/natureneuroscience 1371

Resource NATURe NeURoscIence

a e Verb generation Action observation Theory of mind Visual search

MDTB
Set A Set B
29 task conditions 32 task conditions

Nature movie
Landscape movie
No-go
Theory of mind Animated movie
Go
Action observation Spatial map–easy
Math
Video knots Spatial map–medium
Digit judgment
Finger simple Spatial map–hard Object 2-back Sad faces Finger sequence Rest
Unpleasant scenes
Finger sequence Mental rotation–easy
Pleasant scenes
Object 2-back– Mental rotation–medium
Object viewing
Object 2-back+ Mental rotation–hard
Sad faces
Visual search–easy Response alt.–easy
Happy faces 0.1
Visual search–medium Response alt.–medium
Interval timing

Arbitrary unit
Visual search–hard Response alt.–hard
Motor imagery
Spatial imagery Biological motion
Stroop incongruent
Verb generation Scrambled motion
Stroop congruent
Word reading CPRO
Verbal 2-back–
Rest True prediction Animated movie Landscape movie Finger simple Spatial imagery
Verbal 2-back+ –0.1
Violated prediction
Scrambled prediction

b Instruction Instruction
Task 1 Task 7

c Left-hand presses Right-hand presses Saccades d Reliability Biological motion Mental rotation Spatial map Math

0.6

Reliability
0

Fig. 1 | MDTB. a, Experimental design. A total of four fMRI scanning sessions were collected on the same set of participants using two tasks sets. Each
set consisted of 17 tasks, with 8 tasks in common. The tasks were modeled as 29 task conditions in set A and as 32 in set B, with 14 task conditions
common across both task sets. b, Timing of each task: 5 s instruction period followed by 30 s of task execution. Tasks consisted of a different number
of task conditions (gray bars, range 1–3). c, Unthresholded, group-averaged motor feature maps, displayed on a surface-based representation of the
cerebellar cortex14. d, Across-session reliability of activation patterns for each voxel. e, Group-averaged activation maps for selected tasks, corrected for
motor features. The red-to-yellow colors indicate increases and the blue colors denote decreases in activation, relative to the mean activation across all
conditions. Activity is normalized by the root-mean-square error of the time series fit for each voxel. CPRO, concrete permuted rules operations.

was performed once per imaging run for a 35 s block (Fig. 1b). This cerebellum for movement, they demonstrate that a broad task-
ensured that all tasks were measured against a common baseline, based approach without tightly matched control conditions pro-
allowing for any between-task comparison. To make this approach vides a powerful means of revealing functional organization.
feasible, the tasks were split into two sets (Fig. 1a), and each task set
was tested in two separate fMRI scanning sessions, resulting in a MDTB elicits varied activation patterns across the cerebellum.
total of approximately 5.5 h of functional data per participant. We then characterized the task-related activation patterns that
could not be explained by basic motor features. Overall, we could
Identification of motor features from the MDTB. As a first step, elicit strong and distinguishable patterns of activation (Fig. 1e and
we sought to identify cerebellar regions where the hemodynamic Supplementary Fig. 1) across the cerebellar cortex. To determine
response was closely tied to motor function, specifically hand and the reliability of the activation patterns, we calculated the correla-
eye movements. Our experimental design did not include specific tion of the individual, unsmoothed task activation profiles for each
contrasts that isolated each motor component. Instead, we varied voxel across the two sessions of each set. On average, these task
the motor demands across task conditions; for example, the motor activation profiles were reliable (set A: r = 0.43, 95% confidence
sequence task involved approximately 40 left and right finger interval (CI) = 0.39–0.46; set B: r = 0.42, 95% CI = 0.37–0.46; see
responses, the theory of mind tasks two left-hand responses and Supplementary Fig. 2 for individual participants). The resulting
the movie tasks no responses. We then generated a motor feature voxel-wise reliability map (Fig. 1d) confirmed that this was the case
model, which included the number of left- and right-hand responses for the entire cerebellar cortex, with the exception of lobules I–IV.
and the number of saccadic eye movements made per task (see These lobules are associated with foot movements10,13, a feature
Methods). Using regularized regression, we could then estimate the absent from our tasks.
activation across tasks attributable to motor involvement. Qualitatively, the activation patterns elicited by our task sets rep-
Left- and right-hand movements were associated with acti- licated numerous results obtained in previous neuroimaging stud-
vation increases in the two hand motor areas of the cerebellum ies. For example, right-lateralized activation throughout Crus I,
(Fig. 1c), the anterior hand region located on the boundary of lobules Crus II and VIIb was observed with the verb generation task8 while
V and VI, and the inferior region in lobules VIIIb (ref. 10). Saccadic left-lateralized activation throughout Crus I and Crus II was dem-
eye movements elicited activation in vermis VI, consistent with onstrated with the biological motion task. Consistent with previous
the location of the oculomotor vermis in the macaque monkey11. working memory studies14, the n-back tasks activated two distinct
Compared to previous contrast-based human fMRI studies12, which lateral regions of lobules VII. Recent evidence for medial Crus I and
have yielded relatively inconsistent results, our feature-based map- Crus II activation during movie tasks was also corroborated15.
ping approach resulted in an extraordinarily clear localization of eye The task activation maps also demonstrated some insights
movement activation to the oculomotor vermis. While these results that have not been (or not as clearly) reported in the previous lit-
mainly confirm the well-known functional localization within the erature. The rest condition (contrasted against the mean of all

1372 Nature Neuroscience | VOL 22 | AUGUST 2019 | 1371–1378 | www.nature.com/natureneuroscience

NATURe NeURoscIence Resource
a b c
0.6 I–IV
V
10 mm
Within VI
0.5

Voxel-to-voxel correlation
Within
Between
Between
Within 0.4
Crus I

Crus II
0.3
VIIb
0.2 VIIIa
VIIIb
IX
0.1 X
0 35
Spatial distance (mm)

Fig. 2 | DCBC. a, Correlations between all pairs of voxels with the same distance were calculated and averaged depending on whether they were ‘within’
or ‘between’ regions. Voxel pairs were then binned according to spatial distance in the volume (4–35 mm in steps of 5 mm). b, Average cross-validated
correlation (see Methods) as a function of spatial distance for lobular boundaries. The DCBC is defined as the difference in correlation (within-between)
within each distance bin. The error bars show between-participant s.e.m. for n = 24 participants. c, Strength of the boundaries for lobular parcellation, with
the thickness of the black lines indicating the DCBC value.

the other conditions) was associated with bilateral activation in a (that is, silhouette coefficient) to account for spatial distance. Given
mid-hemispheric region in Crus I and II, effectively forming the that the spatial resolution of fMRI is insufficient to cleanly resolve
cerebellar component of the default mode network5. Similar cer- individual folia, the spatial distance was measured in the volume
ebellar regions were strongly activated during the theory of mind14 (see Methods for details).
and movie tasks15. The finger sequence and visual search tasks led We first employed this evaluation method to determine the
to strong activation in cerebellar hand and eye movement-related degree to which functional boundaries follow the major lobular sub-
areas, respectively. Given that these activation maps were corrected divisions5. This is a question of high practical importance given that
for movement-related activity, these results indicate that these areas lobular boundaries are commonly used to define regions of interest
are especially activated during movements with high attentional for interpreting functional activations in the cerebellum. Notably,
demands. Finally, the action observation task elicited activation in a the correlation between voxels within a lobule was not much greater
distinct set of areas surrounding the motor areas of the cerebellum, than the correlation between voxels that spanned a lobular bound-
especially in the posterior motor representation. When using a dis- ary (Fig. 2b). The correlations, averaged over distances of 4–35 mm,
similarity measure to construct a representational space for all tasks were r = 0.28 (95% CI, 0.26–0.30) within lobules and r = 0.25 (95%
(Supplementary Fig. 4), the action observation condition emerged CI, 0.05–0.46) between lobules. While statistically significant
as one of the most unique activity patterns. (t23 = 4.62, P < .01), the difference was very small (DCBC = 0.03).
The passive picture viewing tasks (that is, sad faces) did not elicit Thus, lobular boundaries do not reflect strong functional subdivi-
much activation in the cerebellum. This is generally consistent with sions in the cerebellum.
the notion that the cerebellum does not receive cortico-pontine pro- The DCBC can also be used to evaluate the strength of individ-
jections from the inferior temporal cortex, a pathway involved in ual boundaries (Fig. 2c). For example, the superior posterior fissure
visual object and scene recognition. To quantify this observation, we separating lobule VI from VII was the strongest lobular boundary
tested the activation patterns of all possible task conditions against (DCBC = 0.152), while the primary fissure, which serves as the
each other. While over 95% of the pairwise comparisons were sig- first principal subdivision of the cerebellum, was relatively weak
nificant (uncorrected P < 0.001), the most notable exceptions were (DCBC = 0.068). The boundary separating Crus I and Crus II did
pairs of the picture viewing tasks (Supplementary Fig. 4a). In con- not predict any functional specialization (DCBC = 0). In sum, many
trast, passively watching engaging movie snippets (nature movie, cerebellar fissures did not demarcate a change in function.
animated movie) resulted in reliable and specific activity patterns
(Fig. 1e and Supplementary Fig. 1), probably related to processes MDTB parcellation uncovers strong functional boundaries. Next,
required for action perception and social cognition. we asked whether a parcellation based on the MDTB data would
more clearly identify functional boundaries. We first estimated a
Cerebellar lobules do not reflect functional subdivisions. One group-based parcellation using all of the MDTB data. Using con-
way to summarize these activation patterns is to subdivide the cer- vex semi-non-negative vector factorization, we decomposed the N
ebellum into functionally distinct regions. However, this approach (tasks) × P (voxels) data matrix into a product of an N × Q (regions)
is only meaningful if there are stable functional subdivisions in the matrix of task profiles and a Q × P matrix of voxel weights. The voxel
cerebellum that generalize across tasks. To address this fundamental weights, but not the task profiles, were constrained to be non-nega-
question, we developed an evaluation metric, which we refer to as tive. Using a winner-takes-all approach, we then assigned each voxel
the distance-controlled boundary coefficient (DCBC). If a bound- to the region with the highest weight. Figure 3b shows the result-
ary divides two functionally heterogeneous regions, then any equi- ing parcellation using ten regions. For this parcellation, the average
distant pair of voxels within a region should have activation profiles DCBC was 0.159 (Fig. 3a, dashed line, t23 = 31.85, P < 1 × 10−10), a
that are more correlated with each other than two voxels that are value higher than that obtained for the strongest lobular boundary.
separated by the boundary (Fig. 2a; see Methods). Specifically, we However, functional parcellations will invariably yield boundar-
calculated correlations between voxel pairs using a range of spatial ies for a given task set, since training and evaluation data overlap.
bins (4–35 mm). The difference between the within- and between- Critically, a good parcellation should be able to predict boundaries
region correlations for each spatial bin then served as our evalu- for a new set of tasks. Therefore, we determined the parcellation
ation criterion. This method extends standard clustering metrics based on all task conditions from set A and evaluated the boundaries

Nature Neuroscience | VOL 22 | AUGUST 2019 | 1371–1378 | www.nature.com/natureneuroscience 1373

Resource NATURe NeURoscIence

a b c
1
0.5

Consistency
Voxel-to-voxel correlation
0.8
0.4
0.6
0.3
0.4
0.2
d
0.1 Within
Between
0
0 35
Spatial distance (mm)

e f g
0.16 0.6
0.2
Individual
0.14 Group
0.5
Voxel-to-voxel correlation

0.12 0.16
MDTB: 7
MDTB: 10 0.4
0.1
MDTB: 17 0.12
DCBC

DCBC
0.08 0.3

0.06 0.08
0.2

0.04
0.1 0.04 Individual
0.02 Group

0
0 0
0 35 0 35
ll

2
>2
ra

0.

5–

1.

5–

Spatial distance (mm) Spatial distance (mm)

ve

0–

1–
0.

1.
O

Cycles/cm

Fig. 3 | MDTB parcellation reveals functional boundaries in the cerebellar cortex. a, Average cross-validated correlations (see Methods) for ‘within’ (red)
and ‘between’ (black) voxel pairs for MDTB parcellation (10 regions). The solid lines indicate the values for the cross-validated estimates; the dashed lines
are the estimates for the full parcellation. b, Ten-region MDTB parcellation. The DCBC for each boundary is visualized by the thickness of the black lines.
c, Proportion of samples in the bootstrapped analysis (participants) where the voxel was assigned to the same compartment as in the original parcellation.
Most voxels had a consistency of assignment >0.6. d, Visualization of boundary uncertainty, using the color scheme in b, but adjusted so that the degree of
transparency is indicative of the uncertainty of the assignment. Voxels that were assigned to a single compartment on less than 50% of the cases are shown
in gray. e, DCBC as a function of the spatial distance for the lower bound of the three MDTB parcellations (colored lines) and lobular parcellation (black line).
f, Within-subject (black) and between-subject (red) reliability of activation patterns overall and across different spatial frequencies. g, DCBC as a function
of voxel distance for the ten-region group parcellation (red) and the average of the ten-region individual parcellations (black). Only the cross-validated
estimates of the prediction performance for novel tasks are shown. In all panels, the error bars show the between-participant s.e.m. for n = 24 participants.

using the unique tasks from set B. We repeated this out-of-sample the cerebellum. Thus, the exact choice of a ‘final’ parcellation is con-
generalization test in the other direction and averaged the two val- strained by practical considerations. In this study, we focus on the
ues. Using this approach, the average DCBC was 0.130, only slightly ten-region parcellation because it provides a useful level of resolu-
lower that the non-cross-validated estimate (Fig. 3a; t23 = 24.232, tion for a full functional characterization.
P < 1 × 10−10). The cross-validated DCBC will underestimate the To assess the stability of the parcellation, we conducted a boot-
true predictive power of the full parcellation, with true performance strap analysis, across both participants and task conditions (for the
on a novel task probably falling between the cross-validated and details, see Methods). The mean Rand coefficient between each of
non-cross-validated DCBC. To remain conservative, we only report the new parcellations and the original parcellation was 0.646 (95%
the cross-validated DCBC estimates for the remainder of the article. CI = 0.55–0.73) for the bootstrap across participants and 0.654 (95%
The exact form of a parcellation depends on the specified num- CI = 0.58–0.73) across task conditions. To quantify the uncertainty
ber of regions. We also derived a parcellation with 7 (Supplementary of specific boundaries, we calculated the proportion of bootstrap
Fig. 5d) or 17 regions (Supplementary Fig. 5f). While the 7-region samples for which each voxel was assigned to the same compart-
parcellation performed slightly worse than the 10-region parcella- ment as in the original parcellation (Fig. 3c). Overall, consistency
tion (DCBC = 0.121, t23 = −4.18, P = 0.00036), the 17- and 10-region was good for most of the cerebellum (Fig. 3d). Lobules I–IV had
parcellations performed comparably (DCBC = 0.133, t23 = 1.57, higher uncertainty, probably a consequence of a lack of foot move-
P = 0.131; Fig. 3e). While there was reasonable agreement across the ments in our task battery.
different MDTB parcellations (Supplementary Fig. 5h), some differ- The parcellations described earlier were based on group data.
ences in the functional subdivisions for the different parcellations To quantify the variability in functional organization across indi-
emerged. While our results clearly show that the MDTB parcella- viduals, we compared the correlation between the task activation
tions reflect true functional boundaries in the cerebellum, they also maps across participants to the within-participant reliability across
make clear that there are a number of equivalent ways to subdivide the two sessions (Fig. 3f). Overall, 27.7% of the pattern variance

1374 Nature Neuroscience | VOL 22 | AUGUST 2019 | 1371–1378 | www.nature.com/natureneuroscience

NATURe NeURoscIence Resource
a d f
MDTB: 10
Task-free: 7 0.16 Adjusted Rand coefficient
0.8
1. MDTB: 7
0.14
2. MDTB: 10 0.6
0.12
3. MDTB: 17
0.4
0.1 Task-free: 7 4. Task-free: 7

DCBC
Task-free: 10
Task-free: 17 5. Task-free: 10 0.2
0.08
6. Task-free: 17
0
0.06 1 2 3 4 5 6
b Task-free: 10 0.04 g MDTB–Task-free correspondence
1.2
0.02
1.0
0 0.8
0 35
Spatial distance (mm) 0.6
0.4
e 0.06
0.2

0.05 0

c Task-free: 17
0.04 h 0.2 MDTB: 10
MDTB: 10
DCBC

0.03

DCBC
0.02 0.1
Task-free: 7
0.01 Task-free: 7 Task-free: 10
Task-free: 10 Task-free: 17
Task-free: 17
0 0
0 35 0 35
Spatial distance (mm) Spatial distance (mm)

Fig. 4 | Comparison of MDTB and task-free parcellations of the cerebellum. a–c, Cerebellar parcellations based on task-free data with 7, 10 and 17
regions6. Thickness of the black lines indicates the DCBC for the corresponding boundary. d, DCBC for different spatial distances for lobular (black),
task-free (dark and light blue) and MDTB (green) parcellations. MDTB parcellation was evaluated in a cross-validated fashion (see text). e, Evaluation
of the same parcellations on task-based data from 186 HCP participants. f, Matrix of adjusted Rand coefficients between three versions of the MDTB
parcellations (7, 10, 17) and the three task-free parcellations (7, 10, 17). g, Correspondence between MDTB and task-free parcellations. A value of 1 (yellow-
green) indicates that the adjusted Rand coefficient between task-free and MDTB parcellations is the same size as the adjusted Rand coefficient between
MDTB parcellations. Lower values (blue-green) indicate weaker agreement between task-free and MDTB parcellations compared to MDTB on its own. h,
Evaluation of MDTB parcellation (derived only from set A) and the task-free parcellations on the three movie tasks from set B. d,e,h, The error bars show
the between-participant s.e.m. for n = 24 or n = 186 (e) participants.

was shared between individuals, whereas 72.3% reflected idiosyn- cating more stability across the MDTB parcellations. The average
cratic patterns. A spatial frequency decomposition of the patterns adjusted Rand index between the MDTB and task-free parcellations
(see Methods) revealed that commonalities across participants was 0.15, indicating that there are systematic differences between
were restricted to the low spatial frequencies (<1 cycle cm−2; activa- the two approaches. To determine where task-free and MDTB par-
tions of more than 5 mm in size), while the fine-grained patterns cellations diverge, we conducted a searchlight analysis, computing
were purely idiosyncratic to the participant. Indeed, a parcellation the adjusted Rand index locally using a 1-cm radius sphere for each
derived from the functional data from the individual significantly pair of parcellations. The results demonstrated that task-free and
outperformed group parcellation in predicting functional bound- MDTB parcellations corresponded most tightly in the mid-lateral
aries for new tasks for that same individual (Fig. 3g; t23 = 5.88, areas of lobule VII. In these ‘default mode’ regions, the agreement
P < 1 × 10−5). between the MDTB and task-free maps (Fig. 4g) was similar to
In summary, using the MDTB data, we were able for the first time the agreement between MDTB maps (Supplementary Fig. 5h). In
to quantitatively demonstrate the existence of distinct functional contrast, in more lateral aspects of lobule VII, and especially areas
regions in the human cerebellum. Our results clearly advocate the engaged in motor control or action observation, the correspondence
adoption of a functional parcellation to replace lobular subdivisions between task-free and MDTB parcellations was much weaker. This
as a tool to summarize functional cerebellar data. is probably due to the relatively low consistency among the task-free
parcellations (Supplementary Fig. 5g).
Task-free parcellations identify overlapping but weaker bound- We then evaluated whether task-free parcellations could predict
aries. Prior work has leveraged the correlational structure of task- functional boundaries in our MDTB data. The average DCBC for
free (or ‘resting state’) fMRI data to derive various parcellations of the task-free 7-, 10- and 17-region parcellations was 0.109, 0.106
the cerebellum, using 75, 106 or 175 regions (Fig. 4a–c). These par- and 0.097 respectively, substantially higher than the lobular par-
cellations were only moderately consistent with each other (Fig. 4f), cellation (t23 = 16.849, P < 1 × 10−10; Fig. 4d). Thus, all of the task-
with an average adjusted Rand index of 0.33 (0 = no communality; free parcellations are, to some degree, able to predict functional
1 = perfect match). Correspondence between the different MDTB boundaries. However, the average task-free DCBC was significantly
parcellations was slightly higher (adjusted Rand index = 0.47), indi- lower than the ‘lower bound’ for our MDTB 10-region parcellation

Nature Neuroscience | VOL 22 | AUGUST 2019 | 1371–1378 | www.nature.com/natureneuroscience 1375

Resource NATURe NeURoscIence

Motor planning Motor planning

1 Left-hand presses 1 2 2 Right-hand presses
Interference resolution Divided attention

Visual working memory Divided attention

3 Saccades 5 6 4 Action observation
Visual letter recognition 10 Motor planning
3
Active maintenance Active maintenance
5 7 8 9
Divided attention 6 Divided attention
Mental arithmetic Verbal fluency

Emotion processing Language processing

7 Narrative 8 Word comprehension
Language processing Narrative

Word comprehension 4 Visual letter recognition

9 Verbal fluency 10 Autobiographical recall
Mental arithmetic Interference resolution

Fig. 5 | Cognitive descriptors for the ten functional regions in the MDTB parcellation. The three features that best characterize each region are listed. The
font size indicates the strength of these feature weights.

(t23 = 5.585, P < 1 × 10−5). This indicates that the MDTB parcellation The dominant features describing the three motor regions
outperformed the task-free parcellations in predicting functional (regions 1, 2 and 3) were left-hand, right-hand and saccadic eye
boundaries on a novel set of task conditions. movements, respectively. The posterior associative motor region
Although the task conditions used for evaluation did not over- (region 4) was driven predominantly by action observation. For the
lap with the tasks used for deriving the MDTB parcellation (see remaining regions, the dominant features related to a range of cog-
also Supplementary Fig. 6), we wanted to ensure that the superior nitive processes. Regions 5 and 6 in the mid-hemispheric aspects of
performance of the MDTB parcellation would generalize to a com- Crus I/II, lateralized to the left and right hemisphere respectively,
pletely separate data set. To this end, we evaluated the MDTB and were associated with attention- and working memory-related fea-
task-free parcellations using data from 186 participants from the tures such as divided attention and active maintenance.
task-based Human Connectome Project (HCP; Fig. 4e) (ref. 16). More medially in both hemispheres were regions 7 and 8, best
Again, the MDTB 10-region parcellation significantly outper- described by features associated with narrative (region 7) and word
formed the three task-free parcellations (7-region: t185 = 22.671, comprehension (region 8). Activity in right-hemispheric region 9,
P < 1 × 10−10; 10-region: t185 = 13.266, P < 1 × 10−10; 17-region: lateral to region 8, was best explained by features related to language
t185 = 28.09, P < 1 × 10−10). processing (for example, verbal fluency and word comprehension).
To ensure that the higher predictive power of the MDTB par- Finally, region 10, encompassing the most lateral aspects of Crus
cellation was not solely driven by regions associated with motor I/II was dominated by autobiographical recall. This region shows
control, we reevaluated the DCBC using only the three movie tasks strong task-free correlations with the frontal pole and other areas
(Fig. 4h). Even though these conditions did not demand any overt related to the default mode network5. Overall, activity in the larger
movement, the advantage of the MDTB over the 7-region (t23 = 2.7, proportion of the cerebellum was explained by features related to
P = 0.01), 10-region (t23 = 5.3, P < 1 × 10−5) and 17-region (t23 = 5.1, cognitive, rather than motor, processes19.
P < 1 × 10−5) task-free parcellation remained significant. Overall,
these results demonstrate that the advantage of the MDTB over Discussion
task-free parcellations extends to new data sets and to conditions Summary. The aim of this study was to derive a comprehensive pic-
that do not involve active tasks. ture of the functional organization of the human cerebellum. To do
this, a group of participants was scanned over the course of four
Characterizing activation by cognitive features. An important fMRI sessions while performing a diverse MDTB. Task-evoked acti-
advantage of a task-based approach is that we can make infer- vation patterns were leveraged to derive a functional parcellation of
ences about the processes that activate the cerebellar cortex. To the cerebellar cortex. Using a new technique to quantitatively evalu-
characterize the functional profiles in each of the regions across ate functional boundaries, we showed that the MDTB parcellation
tasks, we used predefined and non-orthogonal features17. We successfully predicted functional boundaries when tested with a
already successfully applied this approach when characterizing novel set of tasks, outperforming existing parcellation based either
the activation patterns elicited by motor features (Fig. 1c), which on task-free fMRI data or on lobular structure.
could be directly operationalized as the number of finger and
eye movements. To extend this approach, we needed to describe Parcellations of the cerebellar cortex. The lobular architecture of
each task condition in terms of its underlying cognitive features. the cerebellum has provided, both in neurophysiological and neu-
Therefore, we turned to the Cognitive Atlas, an online cognitive roimaging studies, the primary reference for defining subregions4,20.
ontology18, which summarizes the current consensus in cognitive The macroanatomical folding into ten lobules is well conserved
science of the processes associated with a large array of tasks. To across species4 and under strong genetic control21. While the results
construct a feature space, each of the task conditions was rated on from electrophysiological22 and neuroimaging studies5,7 have sug-
each of the cognitive concepts (see Methods). We then estimated gested that lobular boundaries do not demarcate functional sub-
the feature weights for each region using non-negative regres- divisions, we present in this study the first quantitative evaluation
sion. For visualization purposes, we depicted the top three feature of this hypothesis. Indeed, lobular boundaries appear to constitute
weights for each region (Fig. 5). only very weak boundaries in terms of functional organization.

1376 Nature Neuroscience | VOL 22 | AUGUST 2019 | 1371–1378 | www.nature.com/natureneuroscience

NATURe NeURoscIence Resource
The identified functional regions often spanned multiple lobules, lation in predicting functional boundaries in that individual. Of
with many of the boundaries traversing the cerebellar cortex along course, individual parcellations require data collection for each par-
the parasagittal axis. The clear dissociation of anatomical and func- ticipant using at least a subset of our task battery. It is worth noting
tional organization of the cerebellum, as revealed in the present that each individual parcellation was obtained on almost 3 h of data
study, questions the value of summarizing functional and anatomi- per participant, an amount of scanning time that is usually not feasi-
cal data in terms of lobular regions of interest. ble or practical. In future studies, we aim to determine the required
As an alternative, we employed our task-related data to develop amount of data per participant, identify the best subset of tasks and
a parcellation that could comprehensively describe the functional explore the possibility of combining individual and group data to
organization of the cerebellar cortex. Critically, the group-based derive an optimal parcellation.
MDTB parcellation predicted functional boundaries for new tasks
in the same data set, as well as for a completely separate data set Insights about functional topography. An additional advantage of
(HCP task data). These findings provide a compelling demonstra- a multi-domain task-based approach for mapping the cerebellum is
tion of discontinuities in the functional specialization across the that we can not only identify functional boundaries, but relate the
cerebellar cortex. Evidence from meta-analyses has suggested the activation patterns to the task requirements. For many of the tasks
existence of ‘motor’, ‘cognitive’ and ‘affective’ regions of the cerebel- in our battery, the activation patterns were in accord with the results
lum8. However, it has also been suggested that functional variation obtained in previous fMRI studies that have examined a single or lim-
across the cerebellar cortex may be best understood in terms of ited set of task domains. Examples in the present study include work-
smooth gradients7, without definable boundaries. If this were the ing memory, hand movement, language and theory of mind tasks.
case, our DCBC measure, reflecting the difference of within-region By using a rich MDTB, we also identified functional regions that
to between-region correlations, would have been near zero when had not been observed or well described in previous work. A large and
tested on a novel task set. Instead, the values were positive, provid- extended region of the cerebellar cortex was activated during action
ing a rigorous demonstration of functional boundaries in the cer- observation (region 4), surrounding the anterior, but to a much larger
ebellar cortex. extent, posterior hand motor region. Interestingly, the action observa-
An open question is whether the boundaries defined through tion region was also activated during complex movement, as shown
our task-based approach relate systematically to anatomical features by the sequence production task. Taken together, these results suggest
of the cerebellum identified by molecular techniques22. Specifically, that anterior motor regions are more related to primary action execu-
studies investigating aldolase C (zebrin II) expression in Purkinje tion, whereas posterior motor regions are more akin to a ‘premotor’
cells in the rodent23 and primate brain24 have revealed a series of area, perhaps associated more with action planning and action com-
parasagittal zones across the cerebellar cortex. Olivocerebellar pro- prehension. Notably, lesions limited to the posterior cerebellum rarely
jections respect this zonal organization, with single climbing fiber lead to lasting symptoms of ataxia28.
inputs synapsing onto Purkinje cells that lie within a zone23. The A second example comes from our motor feature model, which
organization of zebrin II zones remains to be established in the revealed a region around vermis VI that was strongly associated with
human cerebellar cortex. However, we suspect that the alignment saccadic eye movements (region 3). This finding is consistent with
with the organization observed in this study may not be very tight neurophysiological data from non-human primates showing that this
given that the cerebellar hemodynamic signal is primarily reflec- region is associated with oculomotor control11. However, prior neuro-
tive of mossy fiber input25. Relative to climbing fibers, mossy fiber imaging studies of the cerebellum have proven controversial regarding
innervation patterns are likely to be patchier and more diffuse, this issue. Some studies have also linked this area with eye move-
potentially spanning multiple zonal regions26. ments12, but other studies have argued for a functional role of this
The boundaries identified from task-free fMRI data were also region in more complex cognitive and/or affective processes7. Based
able to predict task-based discontinuities. This finding is in accord on pilot work for this study and other unpublished observations, we
with similar analyses of the cerebral cortex, demonstrating that task- have found it difficult to elicit any cerebellar activation with a simple
based activation patterns in the neocortex can be predicted to some saccadic eye movement task. In contrast, we observed robust activa-
degree by parcellations obtained from the spontaneous fluctuations tion in this area during the visual search task, even when accounting
in the fMRI signal during rest27. However, our MDTB parcella- for the average number of saccades made during a 30 s block. Thus,
tion outperformed alternative task-free parcellations5,6 in predict- our results offer a fresh perspective on the functional role of this
ing functional boundaries for completely different tasks within the region, indicating that the hemodynamic signal is driven by eye move-
MDTB and HCP data sets. While the MDTB parcellation was based ments performed under high attentional demands.
on fewer participants than the other parcellations (24 versus 1,000), The identification of this oculomotor region is also of interest
our data set entailed considerably longer scanning time per partici- given that prior studies have suggested that vermal activation in
pant. One notable difference between the task-free parcellations and lobule VI is associated with emotional processing7. Conditions in
our MDTB parcellation was that the latter assigned homologous our battery designed to engage emotional and affective processing
areas in the left and right hemispheres to different regions. This was (for example, static images of unpleasant and pleasant scenes, sad
the case for regions associated with hand movements (regions 1 and and happy faces) only weakly activated this region. Given the strong
2), working memory (regions 5 and 6) and narrative comprehen- association of this region with eye movements, it may be that prior
sion (regions 7 and 8). While the parcellation suggests some degree activations were more related to differences in saccadic eye move-
of hemispheric asymmetry within the cerebellum, the task activity ments between conditions, rather than the emotional and affective
profiles between homologous regions also shared many similarities processing demands of the tasks. Further support for this hypoth-
(Supplementary Fig. 7). Second, the MDTB parcellation also indi- esis comes from task-free fMRI studies showing that the oculomo-
cated that the areas correlated with the default mode network in tor vermis is functionally connected to visual regions of the cerebral
task-free data could be subdivided into regions related to narrative cortex5,7. An obvious challenge for future research is to explore how
comprehension (regions 7 and 8), language functions (regions 8 and variation in eye movements, affective processing and attentional
9) and autobiographical recall (region 10). demands interact in driving activation within this region.
While our group-based map could predict functional boundaries
in individual participants, the finer spatial details of the functional Conclusions
organization were idiosyncratic for each individual. Consistent with This article presents a comprehensive MDTB for the human cerebel-
this, the individual parcellation outperformed the group parcel- lum, which is unique in its functional diversity and amount of data

Nature Neuroscience | VOL 22 | AUGUST 2019 | 1371–1378 | www.nature.com/natureneuroscience 1377

Resource NATURe NeURoscIence

per individual. The group and individual task contrast maps and the 15. Nguyen, V. T. et al. Distinct cerebellar contributions to cognitive-
group parcellations can be viewed at diedrichsenlab.org/imaging/ perceptual dynamics during natural viewing. Cereb. Cortex 27,
5652–5662 (2017).
mdtb.htm. We anticipate this resource will be useful for two impor- 16. Barch, D. M. et al. Function in the human connectome: task-fMRI and
tant applications. First, the data, combined with our evaluation individual differences in behavior. Neuroimage 80, 169–189 (2013).
criterion, provide a quantitative assessment of functionally defined 17. Diedrichsen, J. & Kriegeskorte, N. Representational models: a common
boundaries, something that has been absent in prior studies. Given framework for understanding encoding, pattern-component, and
representational-similarity analysis. PLoS Comput. Biol. 13, e1005508 (2017).
that our acquisition parameters covered the entire brain, the meth-
18. Poldrack, R. A. et al. The Cognitive Atlas: toward a knowledge foundation for
ods presented in this study can be used to evaluate parcellations of cognitive neuroscience. Front. Neuroinform. 5, 17 (2011).
the neocortex and other brain structures. Second, the novel MDTB 19. Marek, S. et al. Spatial and temporal organization of the individual human
parcellations provide an important tool to define functional regions cerebellum. Neuron 100, 977–993.e7 (2018).
of the cerebellum. Our analyses clearly show that the MDTB parcel- 20. Diedrichsen, J., Balsters, J. H., Flavell, J., Cussans, E. & Ramnani, N. A
probabilistic MR atlas of the human cerebellum. Neuroimage 46, 39–46
lations predict functional boundaries in a novel set of tasks better (2009).
than existing task-free parcellations. Moreover, each region can be 21. Airey, D. C., Lu, L. & Williams, R. W. Genetic control of the mouse
characterized by a rich functional task profile, allowing for a charac- cerebellum: identification of quantitative trait loci modulating size and
terization of the associated cognitive processes. For future research, architecture. J. Neurosci. 21, 5099–5109 (2001).
the parcellation and associated features can provide a useful guide 22. Apps, R. & Hawkes, R. Cerebellar cortical organization: a one-map
hypothesis. Nat. Rev. Neurosci. 10, 670–681 (2009).
in designing studies to test specific functional hypotheses and pro- 23. Sugihara, I. & Shinoda, Y. Molecular, topographic, and functional
vide a reference for interpreting the results. The MDTB functional organization of the cerebellar cortex: a study with combined aldolase C and
parcellation should also be of considerable utility for translational olivocerebellar labeling. J. Neurosci. 24, 8771–8785 (2004).
work, given the hypothesized involvement of cerebellar dysfunction 24. Leclerc, N., Doré, L., Parent, A. & Hawkes, R. The compartmentalization of
in a range of neurological and psychiatric disorders29. A function- the monkey and rat cerebellar cortex: zebrin I and cytochrome oxidase. Brain
Res. 506, 70–78 (1990).
ally defined parcellation can help reveal dysfunction in specific cer- 25. Lauritzen, M. Relationship of spikes, synaptic activity, and local
ebellar regions and cerebro-cerebellar circuits30, providing further changes of cerebral blood flow. J. Cereb. Blood Flow Metab. 21,
insight into the interaction between the cerebellum and neocortex. 1367–1383 (2001).
26. Hawkes, R., Gallagher, E. & Ozol, K. Blebs in the mouse cerebellar granular
Online content layer as a sign of structural inhomogeneity. 1. Anterior lobe vermis. Acta
Anat. (Basel) 158, 205–214 (1997).
Any methods, additional references, Nature Research reporting 27. Tavor, I. et al. Task-free MRI predicts individual differences in brain activity
summaries, source data, statements of code and data availability and during task performance. Science 352, 216–220 (2016).
associated accession codes are available at https://doi.org/10.1038/ 28. Martin, T. A., Keating, J. G., Goodkin, H. P., Bastian, A. J. & Thach, W. T.
s41593-019-0436-x. Throwing while looking through prisms. I. Focal olivocerebellar lesions
impair adaptation. Brain 119, 1183–1198 (1996).
29. Andreasen, N. C. & Pierson, R. The role of the cerebellum in schizophrenia.
Received: 20 September 2018; Accepted: 22 May 2019;
Biol. Psychiatry 64, 81–88 (2008).
Published online: 8 July 2019 30. Moberget, T. et al. Cerebellar volume and cerebellocerebral structural
covariance in schizophrenia: a multisite mega-analysis of 983 patients and
References 1349 healthy controls. Mol. Psychiatry 23, 1512–1520 (2018).
1. Ivry, R. B. & Baldo, J. V. Is the cerebellum involved in learning and cognition?
Curr. Opin. Neurobiol. 2, 212–216 (1992). Acknowledgements
2. Kelly, R. M. & Strick, P. L. Cerebellar loops with motor cortex and prefrontal This work was supported by the James S. McDonnell Foundation (Scholar award to J.D.),
cortex of a nonhuman primate. J. Neurosci. 23, 8432–8444 (2003). the Canadian Institutes of Health Research (no. PJT 159520 to J.D.), a Platform Support
3. Allen, G., Buxton, R. B., Wong, E. C. & Courchesne, E. Attentional Grant from Brain Canada and the Canada First Research Excellence Fund (BrainsCAN
activation of the cerebellum independent of motor involvement. Science 275, to Western University), the National Science Foundation (no. OAC-1649658 to R.A.P.)
1940–1943 (1997). and the National Institute of Health (nos. NS092079 and NS105839 to R.B.I.). Data from
4. Larsell, O. The development of the cerebellum in man in relation to its the Human Connectome Project were analyzed by the Washington University–University
comparative anatomy. J. Comp. Neurol. 87, 85–129 (1947). of Minnesota Consortium (principal investigators: D. Van Essen and K. Ugurbil; no.
5. Buckner, R. L., Krienen, F. M., Castellanos, A., Diaz, J. C. & Yeo, B. T. T. The 1U54MH091657) funded by the 16 National Institutes of Health (NIH) institutes and
organization of the human cerebellum estimated by intrinsic functional centers that support the NIH Blueprint for Neuroscience Research and by the McDonnell
connectivity. J. Neurophysiol. 106, 2322–2345 (2011). Center for Systems Neuroscience at Washington University. Thanks to J. Walters for
6. Ji, J. L. et al. Mapping the human brain’s cortical-subcortical functional assistance in task annotation.
network organization. Neuroimage 185, 35–57 (2019).
7. Guell, X., Schmahmann, J. D., Gabrieli, J. & Ghosh, S. S. Functional gradients
of the cerebellum. eLife 7, e36652 (2018). Author contributions
8. Stoodley, C. J., Valera, E. M. & Schmahmann, J. D. Functional topography of R.B.I. and J.D. originally conceived of the project. M.K., J.D. and R.B.I. designed the
the cerebellum for motor and cognitive tasks: an fMRI study. Neuroimage 59, study. M.K. and C.R.H.-C. collected the data. M.K. and J.D. performed the analyses.
1560–1570 (2012). R.A.P., M.K., J.D. and R.B.I. annotated the cognitive tasks. M.K., J.D. and R.B.I. wrote the
9. Stoodley, C. J. & Schmahmann, J. D. Functional topography in the human manuscript. All authors discussed the results and contributed to the final manuscript.
cerebellum: a meta-analysis of neuroimaging studies. Neuroimage 44,
489–501 (2009). Competing interests
10. Wiestler, T., McGonigle, D. J. & Diedrichsen, J. Integration of sensory and The authors declare no competing interests.
motor representations of single fingers in the human cerebellum. J.
Neurophysiol. 105, 3042–3053 (2011).
11. Ohtsuka, K. & Noda, H. Discharge properties of Purkinje cells in the
oculomotor vermis during visually guided saccades in the macaque monkey.
Additional information
Supplementary information is available for this paper at https://doi.org/10.1038/
J. Neurophysiol. 74, 1828–1840 (1995).
s41593-019-0436-x.
12. Nitschke, M. F. et al. Activation of cerebellar hemispheres in spatial
memorization of saccadic eye movements: an fMRI study. Hum. Brain Mapp. Reprints and permissions information is available at www.nature.com/reprints.
22, 155–164 (2004). Correspondence and requests for materials should be addressed to J.D.
13. Horak, F. B. & Diener, H. C. Cerebellar control of postural scaling and central
Journal peer review information: Nature Neuroscience thanks Fenna Krienen and the
set in stance. J. Neurophysiol. 72, 479–493 (1994).
other, anonymous, reviewer(s) for their contribution to the peer review of this work.
14. Diedrichsen, J. & Zotow, E. Surface-based display of volume-averaged
cerebellar imaging data. PLoS ONE 10, e0133402 (2015). © The Author(s), under exclusive licence to Springer Nature America, Inc. 2019

1378 Nature Neuroscience | VOL 22 | AUGUST 2019 | 1371–1378 | www.nature.com/natureneuroscience

NATURe NeURoscIence Resource
Methods requiring two-choice discrimination, responses were made with the index or
Participants. All participants gave informed consent under an experimental middle finger of the assigned hand, while responses for tasks requiring four-choice
protocol approved by the institutional review board at Western University. A discrimination were made with the index and middle fingers of both hands. By
total of 31 participants were scanned performing set A; 26 of this original cohort including a motor feature model in our analysis, we could account for the motor
returned to perform set B (mean break between sessions = 166 d; s.d. = 153 d, with requirements across the tasks.
half returning about a year later and the other half having sessions separated by
2–3 weeks). The five participants who did not return for set B were not included in Behavioral training. For each task set, participants completed 3 d of training
the analyses. Two additional participants were excluded from the analyses because before the first scanning session. Training included all of the tasks with the
they failed to complete all 32 scanning runs. Therefore, the final sample for the exception of the rest condition and the three movies (set B). For each set, the
MDTB consisted of 24 healthy individuals (16 women, 8 men; mean age = 23.8 three training sessions took place over the course of 4–7 d (set A: mean number of
years, s.d. = 2.6) with no self-reported history of neurological or psychiatric days = 5.2, s.d. = 3.5; set B: mean number of days = 4.4, s.d. = 1.8).
illness. All participants self-reported as right-handed (Edinburgh Handedness The first day was used to familiarize participants with the requirements for
Inventory > 40). The sample size was chosen to allow for an accurate assessment of each of the 17 tasks. Participants were instructed to carefully read the instructions.
inter-participant variability of the functional organization of the cerebellum (see When ready, they initiated a 35 s training block. The number of training blocks
also Life Science Reporting Summary). differed depending on the perceived level of difficulty of the task. For example, the
two-alternative forced choice picture-based tasks (IAPS affective, IAPS emotional)
Experimental tasks. The experimental tasks included in set A were chosen to were practiced for three blocks, while the stroop task was practiced for seven
engage a wide range of processing domains (cognitive, motor, affective, social), blocks. During this training session, a run consisted of consecutive blocks of the
in many cases drawing on tasks that had previously been shown to engage the same task. Online feedback was provided for response-dependent tasks (green or
cerebellum. While recognizing that our selection process was somewhat arbitrary red squares to indicate correct or incorrect responses, respectively). At the end of
and that the tasks would differ on a number of different dimensions, our main each run, an overall accuracy score was provided concerning performance on the
criterion was to use a large battery that broadly sampled different functional tasks requiring a button response.
domains. A full description of the tasks, along with the accompanying references, is On the second training day, switching between tasks was introduced.
provided in Supplementary Table 1. Participants were given 6 runs of training, with each run composed of 1 block for
Set B included eight tasks that had been included in set A (shared tasks, for each of 11 tasks that required manual responses. As on day 1, the timing for the
example, theory of mind, motor sequence) and nine unique tasks. The shared tasks first four runs was self-paced, with the participants allowed to read the instructions
provided a means to establish a common baseline across the two task sets. This at their own pace before initiating the 30 s block. For the final two practice runs,
enables between-task comparison across task sets, which is done by subtracting the the instruction phase was limited to 5 s, thus introducing the protocol that would
mean activation pattern of the shared tasks from each task set. Only tasks that were be used in the scanner. Training on this day only included tasks that required overt
successful at eliciting activation in the cerebellar cortex in set A were included as responses. On the third training day, participants practiced all 17 tasks in four
shared tasks in set B. For some of the novel tasks, we selected conditions that are 10-minute runs (35 s per task), emulating the protocol to be used in the scanner
thought to assay similar processing domains as in task set A. For example, both sessions.
sets included working memory tasks, but the tasks involved different stimulus This training program ensured that participants were familiar with the
dimensions (for example, verbal working memory in set A and spatial mapping in requirements for each task and had considerable experience in switching between
set B). Other tasks (for example, the naturalistic movie-viewing tasks) were novel tasks. In this manner, we sought to minimize the impact of learning during the
in task set B. scanning sessions. On the third training day, performance was asymptotic, with
participants correct on at least 85% of the trials for all of the tasks (range = 85–98%;
Experimental design. Each set consisted of 17 tasks. In every imaging run, each see Supplementary Fig. 3).
task was performed once for 35 s. The 35 s block was divided into a 5 s instruction
period, where the task name (for example, ‘theory of mind task’), the response Eye tracking. Eye tracking data were recorded on the third training session to
effector (‘Use your LEFT hand’) and the button-to-response assignment (‘1 = false obtain an estimate of saccadic eye movements for each of the tasks. An algorithm
belief, 2 = true belief ’) were presented on the screen. This was followed by a 30 s implemented in the Eyelink Toolbox (v. 1.5.0)31 identified saccadic eye movements
period of continuous task performance. In general, novel stimuli were introduced as events where eye velocity briefly exceeded a threshold of 30 degrees s−1. These
across imaging runs to prevent participants from learning specific stimulus– data, tabulated as the mean number of eye movements per task, were included
response associations. The one exception was the motor imagery task where as a motor feature in the second-level feature model. Eye tracking data from two
participants were required to imagine playing a game of tennis. The number of participants in set A and three participants in set B were not obtained due to
trials within the 30 s block varied from 1 (for example, the movie-viewing and technical problems. However, since eye movement behavior was consistent across
mentalizing tasks) to 30 (for example, go/no-go task). Most tasks involved 10–15 participants, we used group-based estimates.
trials per block. The motivation for testing all tasks within a scanning run, as
opposed to testing one task in each run, was to ensure a common baseline for all Scanning sessions. Participants completed four scanning sessions in total,
tasks, enabling between-task comparisons. two with set A and two with set B. The first scanning session for each set was
Three of the shared tasks (object 2-back, visual search, semantic retrieval) conducted within a few days of the final training session (set A: mean = 2.0 d,
had a rapid, discrete trial structure (15 per block), whereby each unique stimulus s.d. = 1.6 d; set B: mean = 2.2 d, s.d. = 1.7 d) and the second scanning session was
(picture, letter, noun) was presented for 1.6 s, with the response required to be completed no more than 7 d after the first scanning session (set A: mean = 3.1 d,
completed within this window, followed by an intertrial interval (ITI) of 400 ms. s.d. = 2.5 d; set B: mean = 2.7 d, s.d. = 2.3 d). Each scanning session consisted of
Three of the shared tasks had a slower discrete trial structure: motor sequence eight imaging runs (10 min per run). Each of the 17 tasks was presented once for
task (trials = 8; trial duration = 4.6 s; ITI = 400 ms); theory of mind (trials = 2; 35 s in each imaging run, producing 16 independent measurements per task. The
duration = 14.6 s; ITI = 400 ms); and action observation (trials = 2; duration = 14 s; task order was randomized across runs. To reduce order effects within each set, no
ITI = 1 s). The remaining two shared tasks, spatial imagery and rest, did not have a two tasks were presented in the same order in two different runs. The order within
discrete trial structure (duration = 30 s). each run, as well as the order of the runs, was kept constant for all participants.
Of the nine unique tasks in set A, six had the rapid discrete trial structure This procedure was chosen to allow for across-participant analyses on the time
(interval timing, International Affective Picture System (IAPS) affective, series level (results not presented in this article). As noted earlier, when possible,
IAPS emotional, verbal 2-back, motor imagery, stroop, math, passive viewing: novel stimuli were used in each run to reduce the recall of specific stimulus–
trial = 15, duration = 1.6 s, ITI = 400 ms; go/no-go: trials = 30; duration = 800 ms; response associations.
ITI = 200 ms). The math task was comprised of 10 trials (duration = 2.6 s;
ITI = 400 ms). The motor imagery task did not have a discrete trial structure Image acquisition. All fMRI data were acquired on a 3T Siemens Prisma at the
(duration = 30 s). Centre for Functional and Metabolic Mapping at Western University. Whole-brain
Of the nine unique tasks in set B, six had a discrete trial structure: the functional images were acquired using an echo-planar imaging sequence with
prediction, spatial map, and response alternatives tasks entailed 6 trials per multi-band acceleration (factor 3, interleaved) and in-plane acceleration (factor
block (duration = 4.8 s; ITI = 200 ms), the mental rotation task 9 trials per block 2), developed at the Center for Magnetic Resonance Research at the University of
(duration = 3 s; ITI = 300 ms), the biological motion task 10 trials per block Minnesota. Imaging parameters were: repetition time = 1 s; field-of-view = 20.8 cm;
(duration = 3 s; ITI = 0 s) and the concrete permuted rules operations (CPRO) phase encoding direction P to A; 48 slices; 3 mm thickness; in-plane resolution
task 4 trials per block (duration = 7.3 s; ITI = 200 ms). The three movie-viewing 2.5 × 2.5 mm2. Gradient echo field maps were acquired for distortion correction of the
tasks (landscape, animated and nature) did not have a discrete trial structure echo-planar imaging images due to B0 inhomogeneities (repetition time = 0.5 s, field-
(duration = 30 s). of-view = 24 cm, 46 slices with in-plane resolution of 3 × 3 × 3 mm3). We also acquired
online physiological recordings of both heart and respiration during each functional
Hand assignment across tasks. For each task requiring responses, the responses run. For anatomical localization and normalization, a 5 min high-resolution scan of
were made with either the left, right or both hands using four-key button boxes. the whole brain was acquired (magnetization-prepared rapid acquisition gradient
Hand assignment was consistent for sets A and B for the shared tasks. For tasks echo; field-of-view = 15.6 × 24 × 24 cm3, at 1 × 1 × 1 mm3 voxel size).

Nature Neuroscience | www.nature.com/natureneuroscience

Resource NATURe NeURoscIence
Image preprocessing. Data preprocessing was carried out using tools from Reliability of activation patterns. To determine intra-participant reliability across
SPM12 (ref. 32), Caret (v. 5.65)33 and spatially unbiased infratentorial template the entire cerebellar cortex, we calculated the correlation between the average
(SUIT v. 3.3)20, as well as custom-written scripts written in MATLAB 2015b. activation estimates for the first and second session for each task set, separately for
For all participants, the anatomical image was acquired in the first scanning each participant. To obtain an overall reliability, we stacked the 29 (A) or 32 (B)
session. Functional data were realigned for head motion within each session, activation estimates for all cerebellar voxels into a single vector and calculated the
and for different head positions across sessions using the six-parameter rigid Pearson correlation between the two estimates. For Fig. 1e, this analysis was also
body transformation. The mean functional image was then co-registered to the performed for each voxel separately. The group-averaged correlations were then
anatomical image and this transformation was applied to all functional images. visualized on the cerebellar flat map.
No smoothing or group normalization was applied.
Spatial frequency of activation patterns. To determine how much of the variance
General linear model (GLM). A GLM was fitted to the time series of each of the activation patterns was common to the group relative to how much was
voxel separately for each imaging run. The 5 s instruction phase for all tasks was idiosyncratic to the individual participants, we calculated two correlations, one
modeled using a single regressor, but was not included in later analyses. Each between task-activity maps between two sessions for the same participant (as for
task was modeled using a boxcar regressor of 30 s, or a combination of multiple reliability) and the second between sessions of different participants. Correlations
regressors if the block contained sub-conditions. These regressors could be were computed on all gray matter voxels in SUIT space. To determine the spatial
2 boxcar regressors of 15 s each (for example, the verb generation task where scale of these common activation patterns, we decomposed the volume image
one sub-condition is word reading and the other is verb generation), 3 boxcar for each task condition into five spatial frequency bands ranging from 0 to 5
regressors of 10 s each (for example, visual search, display sizes of 4, 8 or 12) or cycles cm−2. This decomposition was done separately for each participant, study,
2 event-related regressors (for example, the stroop task, where each trial was session and task condition. The within- and between-participant correlations were
congruent or incongruent). The rest condition was not modeled explicitly, but then computed for each spatial frequency band.
rather used as an implicit baseline in the model.
The quality of the GLM in modeling the blood-oxygen-level dependent signal Evaluating functional boundaries. We developed a method to evaluate functional
response was determined by measuring the consistency of the activation patterns boundaries from fMRI data. The rationale of the method is that, if a boundary is
in the cerebellum across runs. This measure indicated that it was advantageous to dividing two functionally heterogeneous regions, then two voxels that lie within the
omit the traditional high-pass filtering operation before the linear model (default same region should have more similar functional profiles than two voxels that are
operation in SPM). Instead, we opted to rely on the high-dimensional temporal in different regions (Fig. 2a; equation (1)). Because functional organization tends
autocorrelation model (the FAST option in SPM) to determine the optimal to be smooth, the correlation between two voxels will be higher for two adjacent
filtering, implemented in the GLM estimation. The beta weights from the first-level voxels and fall off as the spatial distance increases10. To control for distance, we
GLM were univariately pre-whitened by dividing them by the square root of the calculated the activation pattern correlations for all pairs of voxels separated by a
residual mean-square image. To include rest as a task condition in all subsequent specific Euclidean distance, using spatial bins ranging from 4 to 35 mm. Of course,
analyses, we added a zero as an estimate for the rest condition and then removed it would have been preferable to measure the distance on the cerebellar cortical
the mean for each voxel across all conditions. As such, the beta weights expressed sheet, rather than in the volume. However, a veridical surface reconstruction of
the amount of activation elicited by each condition relative to the mean of all the cerebellar folia is only possible for resolutions higher than 0.2 mm36. The use of
conditions. volumetric distances will slightly favor lobular boundaries, since two voxels within
To combine activation estimates across the two tasks sets, the mean of the the same lobule will tend to be closer on the cerebellar cortex than two voxels
shared tasks was removed separately for each set. Both sets were then combined, separated by a fissure, even if their distance in the volume is matched. To exclude
retaining the repeated estimates for the shared task. This resulted in a total of spatial correlations that were driven by noise, we used a cross-validated
61 estimates (set A = 29; set B = 32) for the 47 unique conditions. The activation correlation. Using ui,1 to represent the functional profile (zero-meaned) of voxel
patterns were re-centered by removing the overall mean across all 61 conditions. i from one session, and uj,2 the functional profile of voxel j from the other session,
the correlation was calculated as:
Cerebellar spatial normalization. The SUIT toolbox (v.3.3) in SPM12 was used
1
to isolate the cerebellum from the rest of the brain and provide a normalization to 2
∑i, j (uiT,1u j,2 + uiT,2u j,1)
a spatially unbiased template of the cerebellum34. The resulting cerebellar isolation r= (1)
mask was hand-corrected to ensure that it did not contain any shared voxels ∑i, j (uiT,1ui,2) ∑i, j (uTj,1u j,2)
between the superior cerebellum and the directly abutting cerebral cortical regions
of the inferior temporal and occipital cortex. where the sum was done on all voxel pairs i,j in the corresponding 5 mm bin.
The probabilistic maps for the cerebellum were normalized into SUIT space Separate correlations were calculated for voxel pairs from the same region (within)
using the diffeomorphic anatomical registration through exponentiated lie and those where the voxel pairs came from different regions (between). We
algebra algorithm35. This algorithm deforms the cerebellum to simultaneously excluded voxels where the term uiT,1ui,2 was negative, since it indicated the absence
fit the probability maps of cerebellar gray and white matter onto the SUIT atlas of any reliable tuning across the two sessions. The difference between the within-
template. This non-linear deformation was applied to both anatomical and region and between-region correlations defined the DCBC. A positive DCBC value
functional data. The activation estimates (that is, the beta weights) and residual indicates that voxel pairs originating from the same region are more functionally
mean-square images from the first-level GLM were resliced into SUIT space. related than voxel pairs that lie across boundaries. The DCBC was calculated for
All images were masked with the cerebellar mask to avoid activation influences each participant and spatial bin separately, and then averaged.
from the inferior occipital cortex. All data were visualized on a surface-based, The DCBC can serve not only as a global measure of a parcellation (averaging
flat-map representation of the cerebellar cortex in the SUIT toolbox. The flat-map across the cerebellum and spatial bins), but also as a measure to evaluate the
representation allows the spatial extent of task-evoked activation patterns to be strength of individual boundaries. For the latter, we first identified boundaries
fully visualized. Note that this flat map is not a true unfolding of the cerebellar using an edge-based connectivity scheme37. The strength of a given boundary is
cortex, but averages over a substantial number of folia. Therefore, it is meant for defined by the DCBC calculated only on the voxel pairs from the two regions that
display purposes only14. are separated by that boundary. To visualize boundary strength, the thickness of
the boundary on the flat map was based on its DCBC value.
Motor feature model. Our primary goal was to study task-evoked activation We applied this boundary evaluation procedure to MDTB parcellations, as well
patterns in the cerebellum beyond the well-known domain of motor function. as parcellations based on lobular boundaries or task-free fMRI data. The lobular
Although not designed explicitly to measure motor-related activation, the 61 task parcellation was obtained from a probabilistic atlas of the human cerebellum20
conditions differed in the number of manual responses, as well as eye movements. that includes regions for lobules I–IV, V, VI, Crus I, Crus II, VIIb, VIIIa, VIIIa, IX
To account for these motor-related activations, we generated a motor feature model and X. To ensure that that poor performance of the lobular parcellation was not
(task conditions × three motor features). For the hand movements, we entered the due to inaccuracies in detecting the lobular boundaries, we repeated the analysis
number of left- and right-hand presses for each task during the scanner runs. For using a manual lobular parcellation in five participants. The parcellation from this
eye movements, we used the group-averaged eye movement data from the third sample predicted functional boundaries about as well as the one derived from the
training day to estimate the number of saccades for each task condition. All motor probabilistic atlas (DCBC: 0.022 versus 0.025; t5 = 1.441, P = 0.209). The task-free
features were encoded in terms of movements s−1 and z-normalized. 10-region parcellation6 was based on data archived as part of the HCP, while the
To extract and remove the motor-related activation across tasks, the 3 motor other two were based on a large 1,000-person data set collected at Harvard and
features were combined with an indicator matrix that had a 1 for each of the 61 Massachusetts General Hospital5. All parcellations were sampled into SUIT space
task conditions. To estimate and subsequently remove the influence of the motor and evaluated using our MDTB.
features, we estimated the linear model with L2-norm regression (fixed λ of 0.01)
from the beta estimates of each participant (task conditions × voxels × participant). MDTB data set parcellation. To derive a parcellation from the MDTB, we used the
The average of all task conditions was used as a baseline measure and subtracted activation profiles of gray-matter voxels averaged in SUIT volumetric space across
from the motor-corrected activation estimates. The activation estimates for the participants. We used convex non-negative matrix factorization38 to decompose
shared tasks were first averaged; then a group average was computed for the the N (tasks) × P (voxels) data matrix into a product of an N × Q (regions) matrix
purposes of visualization on the cerebellar flat map14. of task profiles and a Q × P matrix of voxel weights. The voxel weights, but not the
Nature Neuroscience | www.nature.com/natureneuroscience
NATURe NeURoscIence Resource
task profiles, are constrained to be non-negative. Furthermore, the task profiles are across the sets, we first subtracted the mean of the shared task conditions from
convex combinations of the raw data. Compared to other decomposition methods, each imaging run. Cross-validation ensures that the average (expected) value of the
such as independent component analysis, this method has the advantage that dissimilarity measure is 0 if the two activation patterns only differ by noise. This
voxels cannot be explained by an inverted or negative regional task profile. This allowed us to test for significant differences between activation patterns using a
constraint is also reasonable given that mossy fiber input, the main neural signal one-sample t-test against 0.
driving the blood-oxygen-level dependent signal response in the cerebellum, is Classical multidimensional scaling was employed to visualize the distances
excitatory25. To ensure convergence, we started the decomposition with random between all possible pairs of task conditions. For the purposes of visualization,
initializations and selected the iteration with the best reconstruction of the original the pairwise distances for the shared tasks were averaged so that there were 47
data. A winner-takes-all approach was adopted to assign each voxel to the region (rather than 61) task conditions in the representational dissimilarity matrix.
with the highest weight. Multidimensional scaling projects the N-dimensional representational dissimilarity
To allow for a direct comparison with existing task-free parcellations5,6, we used matrix into a lower-dimensional space so that distances from the higher space
parcellations with 7, 10 and 17 regions (Supplementary Fig. 5d–f). Parcellations are preserved with as much integrity as possible. Multidimensional scaling was
involving regions within this range achieved similar reconstruction accuracy and performed on the group-averaged representational dissimilarity matrix and the
quality of functional boundaries. first three dimensions were visualized in a three-dimensional space.
We also derived separate parcellations for each participant to determine
whether boundaries could be predicted better using an individual approach. An Feature-based approach. The power of our task-based approach in studying the
advantage of the individual parcellation is that idiosyncrasies of within-participant cerebellum is that we can identify the involvement of each region across functional
organization are captured. The disadvantage is that individual parcellations are domains and different task variations. To summarize the task activation profiles
derived on substantially fewer data than the group. for each region, we used a feature-based encoding model. The features included
the three motor features (see earlier) and cognitive features, selected to capture the
Bootstrap analysis. To obtain a measure of boundary uncertainty, we performed hypothetical mental processes involved in each task. To derive these features, we
bootstrap analyses across both participants and task conditions. For participants, used an online cognitive ontology18, an atlas of tasks and the concepts associated
we repeatedly drew 24 participants (with replacement) from our sample, with those tasks. Of the 815 concepts currently included in the atlas, 46 were
averaged the data and derived a new functional parcellation. To be able to relate judged to provide an appropriate and sufficient characterization of the tasks in our
the parcellations to each other, each parcellation used the original solution as a battery, creating a feature matrix (47 task conditions × 46 features). For example,
starting value. For the task conditions, we repeatedly drew 47 task conditions (with features such as semantic knowledge and lexical processing were associated with
replacement) from our data, again deriving a new parcellation each time. For each tasks such as verb generation and semantic prediction; emotion recognition was
analysis, we repeated this process 100 times. associated with the IAPS emotional processing task and the biological motion
To evaluate the consistency of the parcellations globally, we calculated task. As with the motor feature model, each feature was z-standardized and
the adjusted Rand index, which measures the correspondence between two feature weights for each region were estimated with non-negative regression. For
parcellations (0 = overlap not different from chance, 1 = perfect overlap). For a visualization purposes, the three highest weights for each region were computed.
regionally specific analysis, we counted the number of times that each voxel was
assigned to the same (most frequent) region. For visual display (Fig. 3d), we then Statistical analysis and inference. Unless otherwise noted, all statistical tests
used this assignment certainty to determine the transparency of the region coloring were based on the n = 24 participants of our sample, considering participant as a
(<50% = fully transparent, 100% = fully opaque). random effect. Therefore, all t-tests were repeated measures with 23 d.f. Only the
analysis presented in Fig. 4e is based on the HCP task data and was therefore based
Evaluation of functional parcellations. To evaluate the group and individual on n = 186 participants. All t-tests were two-sided. Data distribution was assumed
MDTB parcellations, we wanted to know how well functional boundaries could to be normal, but this was not formally tested.
be predicted for each participant using a completely novel set of tasks. Because we
did not acquire data with a third, independent task set, we used the existing data to Reporting Summary. Further information on research design is available in the
estimate lower and upper bounds of predictability. For the lower bound, we derived Nature Research Reporting Summary linked to this article.
the parcellation using the data from set A and evaluated it with the data from set
B, using the unique tasks only. This procedure was then repeated with the task sets Data availability
reversed and the results were averaged across the two cross-validation folds. Note The activation maps and functional parcellations are available from http://
that the outcome of this analysis will probably result in a lower value than would www.diedrichsenlab.org/imaging/mdtb.htm. The raw behavioral and imaging
be obtained with the final segmentation, since each parcellation is based on half of data for the cerebellum are also available on the data sharing repository https://
the available data. As such, we used this estimate as an approximate lower bound. openneuro.org/.
We also evaluated a parcellation derived from both sets A and B. We evaluated
this parcellation as before, excluding the shared tasks from both task sets to make
the estimate consistent with the lower-bound estimate. Because there is overlap Code availability
between the data used for training and evaluation, the performance measure is The experimental code is available at https://github.com/maedbhk/MDTB-
overfitted and, therefore, was taken as an approximate upper bound. The true Cerebellum.
performance of the full parcellation, if applied to a completely new task set, would
probably fall between these lower and upper bounds.
The anatomical and task-free parcellations could be directly evaluated on References
the MDTB data since each parcellation was derived from independent data. For 31. Cornelissen, F. W. & Peters, E. M. The Eyelink Toolbox: eye tracking with
consistency, we excluded the data from the shared tasks in the evaluation set. MATLAB and the Psychophysics Toolbox. Behav. Res. Methods Instrum.
To evaluate the degree to which the results depended on the similarity of the Comput. 34, 613–617 (2002).
tasks in the training and test sets, we repeated the analysis, this time selecting the 32. Ashburner, J. et al. SPM12 Manual www.fil.ion.ucl.ac.uk/spm/doc/spm12_
seven most distinct task conditions in each test set (Supplementary Fig. 6). The manual.pdf (2019).
conditions were selected by computing the distance between activity patterns for 33. Van Essen, D. C. Cortical cartography and Caret software. Neuroimage 62,
each test condition to each training condition (Supplementary Fig. 4a). Then, we 757–764 (2012).
identified for each test condition the closest match in the training set and selected 34. Diedrichsen, J. A spatially unbiased atlas template of the human cerebellum.
the seven test conditions for which this closest match was most dissimilar. Neuroimage 33, 127–138 (2006).
To further validate our results, we evaluated the MDTB and task- 35. Ashburner, J. A fast diffeomorphic image registration algorithm. Neuroimage
free parcellations on the task-based data from the HCP data set (https:// 38, 95–113 (2007).
db.humanconnectome.org)16. We used data from the 214 most recently added 36. Sereno, M. I., Diedrichsen, J., Tachrout, M., Silva, G. & De Zeeuw, C.
participants (scanned at 3T). Of the 214, 186 participants had complete data sets Reconstruction and unfolding of the human cerebellar cortex from
and these constituted our final sample. For each participant, we evaluated the high-resolution post-mortem MRI. Society for Neuroscience, abstr.
parcellations on a set of 22 contrast maps from 7 tasks (all against rest). 733 (2014).
37. Oosterhof, N. N., Wiestler, T., Downing, P. E. & Diedrichsen, J. A comparison
Representational structure of task-related activation patterns. Representational of volume-based and surface-based multi-voxel pattern analysis. Neuroimage
similarity analysis17 was used to investigate the representational structure of task- 56, 593–600 (2011).
related activation patterns from the MDTB cerebellar data. The dissimilarity 38. Ding, C. H., Li, T. & Jordan, M. I. Convex and semi-nonnegative matrix
between the motor-corrected activation patterns was measured for each pair of factorizations. IEEE Trans. Pattern Anal. Mach. Intell. 32, 45–55 (2010).
task conditions using the cross-validated Mahalanobis distance, using the imaging 39. Walther, A. et al. Reliability of dissimilarity measures for multi-voxel pattern
runs as independent partitions39. To calculate the distances between conditions analysis. Neuroimage 137, 188–200 (2016).

Nature Neuroscience | www.nature.com/natureneuroscience

 nature research | reporting summary
Corresponding author(s): Diedrichsen, Jörn
Last updated by author(s): May, 22, 2019

Reporting Summary
Nature Research wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency
in reporting. For further information on Nature Research policies, see Authors & Referees and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.
n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement
A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly
The statistical test(s) used AND whether they are one- or two-sided
Only common tests should be described solely by name; describe more complex techniques in the Methods section.

A description of all covariates tested

A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons
A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient)
AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals)

For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted
Give P values as exact values whenever suitable.

For Bayesian analysis, information on the choice of priors and Markov chain Monte Carlo settings
For hierarchical and complex designs, identification of the appropriate level for tests and full reporting of outcomes
Estimates of effect sizes (e.g. Cohen's d, Pearson's r), indicating how they were calculated
Our web collection on statistics for biologists contains articles on many of the points above.

Software and code

Policy information about availability of computer code
Data collection The 26 task were presented and behavioral data collected using custom-written code in MATLAB R2015b. The low-level functionality
from psychtoolbox (3.0.13) was used for display and execution of the tasks tested in both the behavioural and fMRI sessions. The Eyelink
toolbox (MATLAB toolbox) was used to collect eye movement data.

Data analysis SPM12 and SUIT (3.3) toolboxes were used to analyze the data. Caret was used for visual display. Custom code was written in MATLAB
(available here https://github.com/maedbhk/MDTB-Cerebellum)
For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors/reviewers.
We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Research guidelines for submitting code & software for further information.

Data
Policy information about availability of data
All manuscripts must include a data availability statement. This statement should provide the following information, where applicable:
- Accession codes, unique identifiers, or web links for publicly available datasets
- A list of figures that have associated raw data
- A description of any restrictions on data availability
October 2018

The raw data underlying the paper will be uploaded to a data-sharing repository (https://openneuro.org/) for download by
interested members of the scientific community. The aggregated data (individual activation maps and functional parcellations) are available for download at
www.diedrichsenlab.org/imaging/mdtb.htm. The experimental and analysis code
is available on a code-sharing website (https://github.com/maedbhk/MDTB-Cerebellum).

1
 nature research | reporting summary
Field-specific reporting
Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.
Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences
For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

Life sciences study design

All studies must disclose on these points even when the disclosure is negative.
Sample size No formal power analysis was conducted, but the number of participants and the amount of data per participant was determined from prior
experience. For the within-subject sample size, we required 5.5 hours of fMRI data for each participant. This gave us enough power to reliably
estimate activity patterns for the large number of tasks, as well as having enough data to perform out-of-sample prediction tests. A sample
size of n=24 participants (after exclusions) was deemed sufficient to both estimate a representative mean organization, as well as obtaining
reliable estimates of the between-subject variability.
To test generalization of our results, we also utilized the task data from N=186 participants from the human connectome project (https://
www.humanconnectome.org/)

Data exclusions We originally recruited 31 participants. N=5 participants could not return to complete the second task set as they had moved. These data
were not included in the final analyses. Two additional participants were excluded from the analyses as they failed to complete all 32 scanning
runs for technical reasons. Eye-tracking data from two participants in set A and three participants in set B were not obtained due to technical
problems

Replication The data includes a direct replication, as there are two identical sessions for each task set). There was good reliability of activation patterns
across sessions (within a task set). Models were tested using prediction accuracy for a complete separate task set.

Randomization The sequence of task was randomized across imaging runs. All of the participants performed the same sequence of tasks (and the same
number / order of runs) to enable analysis on the timeseries across participants.

Blinding Blinding was not applicable. Participants were not sorted into control and / experimental groups.

Reporting for specific materials, systems and methods

We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material,
system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.

Materials & experimental systems Methods

n/a Involved in the study n/a Involved in the study
Antibodies ChIP-seq
Eukaryotic cell lines Flow cytometry
Palaeontology MRI-based neuroimaging
Animals and other organisms
Human research participants
Clinical data

Human research participants

Policy information about studies involving human research participants
Population characteristics The final sample consisted of 24 healthy, right-handed individuals (16 females, 8 males; mean age=23.8 years old, SD=2.6) with
no self-reported history of neurological or psychiatric illness.

Recruitment Undergraduate and graduate students were recruited (via posters) from the larger student body at Western University. Thus, our
sample was biased towards relatively high-functioning, healthy and young individuals. While we don't expect cerebellar
October 2018

organization to be dramatically different in this group, caution needs to be exercised when generalizing the results to the general
population.

Ethics oversight The Ethics committee at Western University approved all experimental protocols (Protocol number: 107293)

Note that full information on the approval of the study protocol must also be provided in the manuscript.

2
Magnetic resonance imaging

nature research | reporting summary

Experimental design
Design type Task-based

Design specifications Two task sets. 2 fMRI scanning sessions per task set. 8 functional imaging runs per session (10-min each). 17 tasks per
imaging run (35 s each).

Behavioral performance measures Variables recorded: response made, number of correct responses, false alarms, missed responses, response time.
Accuracy (% correct) and reaction time (ms) were collected and averaged across tasks per participant.

Acquisition
Imaging type(s) EPI, MPRAGE, and GRE field maps

Field strength 3T

Sequence & imaging parameters EPI: Gradient echo, multi-band (factor 3, interleaved) with an in-plane acceleration (factor 2). Imaging parameters were:
TR=1 sec, FOV=20.8cm, phase encoding direction was P to A, acquiring 48 slices with in-plane resolution of 2.5 mm x 2.5
mm and 3 mm thickness. For anatomical localization and normalization, a 5-min high-resolution scan of the
whole brain was acquired (MPRAGE, FOV=15.6 cm x 24 cm x 24 cm, at 1x1x1 mm voxel size).

Area of acquisition whole-brain

Diffusion MRI Used Not used

Preprocessing
Preprocessing software Data preprocessing was carried out using tools from SPM 12, Caret, and SUIT, as well as custom written
scripts written in MATLAB 2015b. For all participants, the anatomical image was acquired in
the first scanning session and reoriented to align with the Left-Inferior-Posterior (LPI) coordinate frame.
Functional data were re-aligned for head motion within each session, and for different head positions
across sessions using the 6-parameter rigid body transformation. The mean functional image was then
co-registered to the anatomical image, and this transformation was applied to all functional images. No
smoothing or anatomical normalization was applied to the functional images.

Normalization The probabilistic maps for the cerebellum were normalized into SUIT space using the
diffeomorphic anatomical registration (DARTEL) algorithm. This algorithm deforms the cerebellum to
simultaneously fit the probability maps of cerebellar gray and white matter onto the SUIT atlas template.
This non-linear deformation was applied to both the anatomical and functional data. The activation
estimates (i.e., the beta weights), and residual mean-square images from the first-level GLM were
resliced into SUIT space.

Normalization template The spatially unbiased infratentorial template (SUIT) toolbox (v3.2) in SPM 12 was used to isolate the
cerebellum from the rest of the brain and to provide a normalization to a spatially unbiased template of
the cerebellum. The segmentation procedure was used to create probability maps of gray and white
matter, allowing us to separate cerebellar and cortical tissue. The resulting cerebellar isolation mask was
hand corrected to ensure that it did not contain any shared voxels between the superior cerebellum and
the directly abutting cerebral cortical regions of the inferior temporal and occipital cortex.

Noise and artifact removal The cerebellar isolation mask was hand corrected to ensure that it did not contain any shared voxels between the
superior cerebellum and the directly abutting cerebral cortical regions of the inferior temporal and occipital cortex.

Volume censoring We excluded the first 3 volumes of each imaging run. Otherwise no volumes were censored.

Statistical modeling & inference

Model type and settings Types: multivariate, univariate, RSA, and predictive.

Effect(s) tested Define precise effect in terms of the task or stimulus conditions instead of psychological concepts and indicate whether
ANOVA or factorial designs were used.
October 2018

Specify type of analysis: Whole brain ROI-based Both

Statistic type for inference The main inferences in the paper are all made on multivariate measures that integrated information from the entire
(See Eklund et al. 2016) cerebellum.

Correction Given that we used a single cerebellum-wide evaluation criterion, no correction for the number of voxel-wise or region-
wise tests were necessary.

3
Models & analysis

nature research | reporting summary

n/a Involved in the study
Functional and/or effective connectivity
Graph analysis
Multivariate modeling or predictive analysis

Multivariate modeling and predictive analysis We developed a novel method to evaluate functional boundaries from fMRI data. The rationale of the
method is that, if a boundary is dividing two functionally heterogeneous regions, then two voxels that lie
within the same region should have more similar functional profiles than two voxels that are in different
regions Because functional organization tends to be smooth, the correlation between
two voxels will be higher for two adjacent voxels, and fall off as the spatial distance increases To
control for distance, we calculated the activation pattern correlations for all pairs of voxels separated by
a fixed Euclidean distance, using spatial bins ranging from 4 mm to 35 mm.
To derive a parcellation with the data from our task battery, we used the activation profile averaged in
SUIT volumetric space across participants. The parcellation was based only on data from voxels
assigned to gray matter. As a clustering approach, we used convex semi-nonnegative matrix
factorization39,
which decomposes the N (tasks) x P (voxels) data matrix into a product of an N x Q (regions) matrix of
task profiles and an Q x P matrix of voxel weights. A winner-take-all approach was adopted to assign each
voxel to the region with the highest weight. To ensure convergence, we started the decomposition with
random initializations, and selected the iteration with the best reconstruction of the original data. We
stopped when the current estimate of the best solution was obtained five times without being replaced by
a better solution.
Representational similarity analysis (RSA) was used to investigate the representational structure of
task-related activation patterns from the MDTB dataset in the cerebellum. The dissimilarity between the
motor-corrected activation patterns was measured for each pair of task conditions using the crossvalidated
Mahalanobis distance, using the imaging runs as independent partitions. To calculate the distances
between conditions across the sets, we subtracted the mean of the shared task conditions from
each imaging run first. Cross-validation ensures that the average (expected) value of the dissimilarity
measure is zero if the two activation patterns only differ by noise. This allowed us to test for significant
differences between activation patterns using a one-sample t-test against zero.
To summarize the task activation profiles for each region, we used a feature-based encoding method. The
features included three motor features and cognitive features, selected to capture the hypothetical mental
processes involved in each task. Each feature was z-standardized and feature weights for each region were
estimated with non-negative regression.

October 2018