Homological scaffolds of brain functional

networks
G. Petri1, P. Expert2, F. Turkheimer2, R. Carhart-Harris3, D. Nutt3, P. J. Hellyer4
rsif.royalsocietypublishing.org and F. Vaccarino1,5
1
ISI Foundation, Via Alassio 11/c, 10126 Torino, Italy
2
Centre for Neuroimaging Sciences, Institute of Psychiatry, Kings College London, De Crespigny Park,
London SE5 8AF, UK
3
Centre for Neuropsychopharmacology, Imperial College London, London W12 0NN, UK
Research 4
Computational, Cognitive and Clinical Neuroimaging Laboratory, Division of Brain Sciences, Imperial College
London, London W12 0NN, UK
5
Cite this article: Petri G, Expert P, Turkheimer Dipartimento di Scienze Matematiche, Politecnico di Torino, C.so Duca degli Abruzzi no 24, Torino 10129, Italy
F, Carhart-Harris R, Nutt D, Hellyer PJ,
Networks, as efficient representations of complex systems, have appealed to
Vaccarino F. 2014 Homological scaffolds of
scientists for a long time and now permeate many areas of science, including
brain functional networks. J. R. Soc. Interface neuroimaging (Bullmore and Sporns 2009 Nat. Rev. Neurosci. 10, 186–198.
11: 20140873. (doi:10.1038/nrn2618)). Traditionally, the structure of complex networks has
http://dx.doi.org/10.1098/rsif.2014.0873 been studied through their statistical properties and metrics concerned with
node and link properties, e.g. degree-distribution, node centrality and modular-
ity. Here, we study the characteristics of functional brain networks at the
mesoscopic level from a novel perspective that highlights the role of inhomo-
Received: 5 August 2014 geneities in the fabric of functional connections. This can be done by focusing
Accepted: 3 October 2014 on the features of a set of topological objects—homological cycles—associated
with the weighted functional network. We leverage the detected topological
information to define the homological scaffolds, a new set of objects designed to
represent compactly the homological features of the correlation network and
simultaneously make their homological properties amenable to networks theor-
Subject Areas: etical methods. As a proof of principle, we apply these tools to compare resting-
computational biology state functional brain activity in 15 healthy volunteers after intravenous infusion
of placebo and psilocybin—the main psychoactive component of magic mush-
Keywords: rooms. The results show that the homological structure of the brain’s functional
patterns undergoes a dramatic change post-psilocybin, characterized by the
brain functional networks, fMRI, persistent
appearance of many transient structures of low stability and of a small
homology, psilocybin number of persistent ones that are not observed in the case of placebo.

Author for correspondence:

P. Expert
1. Motivation
The understanding of global brain organization and its large-scale integration
e-mail: [email protected]
remains a challenge for modern neurosciences. Network theory is an elegant frame-
work to approach these questions, thanks to its simplicity and versatility [1]. Indeed,
in recent years, networks have become a prominent tool to analyse and understand
neuroimaging data coming from very diverse sources, such as functional magnetic
resonance imaging (fMRI), electroencephalography and magnetoencephalography
[2,3], also showing potential for clinical applications [4,5].
A natural way of approaching these datasets is to devise a measure of dynami-
cal similarity between the microscopic constituents and interpret it as the strength
of the link between those elements. In the case of brain functional activity, this often
implies the use of similarity measures such as (partial) correlations or coherence
[6–8], which generally yield fully connected, weighted and possibly signed adja-
cency matrices. Despite the fact that most network metrics can be extended to
the weighted case [9–13], the combined effect of complete connectedness and
Electronic supplementary material is available edge weights makes the interpretation of functional networks significantly
harder and motivates the widespread use of ad hoc thresholding methods
at http://dx.doi.org/10.1098/rsif.2014.0873 or
[7,14–18]. However, neglecting weak links incurs the dangers of a trade-off
via http://rsif.royalsocietypublishing.org.

& 2014 The Authors. Published by the Royal Society under the terms of the Creative Commons Attribution
License http://creativecommons.org/licenses/by/4.0/, which permits unrestricted use, provided the original
author and source are credited.
 (a) (b) 2
k k

rsif.royalsocietypublishing.org
d d
j a j a
c c
b b
i i
e e
h f h f
g g
l l

Figure 1. Panels (a,b) display an unweighted network and its clique complex, obtained by promoting cliques to simplices. Simplices can be intuitively thought as

J. R. Soc. Interface 11: 20140873

higher-dimensional interactions between vertices, e.g. as a simplex the clique (b,c,i) corresponds to a filled triangle and not just its sides. The same principle applies
to cliques—thus simplices—of higher order. (Online version in colour.)

Table 1. List of notations.

name symbol definition

graph G a graph G ¼ (V, E) is a representation of a set V of nodes i interconnected by edges or links eij [ E; this
interaction can be weighted, directional and signed
clique c a completely connected subgraph C ¼ (V’, E’) contained in an undirected and unweighted graph
G ¼ (V, E) (V 0 , V, eij [ E 0 8i, j [ E 0 )
k-simplex sk formally, a convex hull of k þ 1 nodes [ p0, p1, . . . ,pk], it is used here as a generalization to higher
dimensions of the concept of link, e.g. a 2-simplex is a triangle, a 3-simplex a tetrahedron. The faces f
of sk are obtained as subsets of [ p0, p1, . . . ,pk]
simplicial complex K a topological space composed by attaching simplices s, with two conditions: (i) if s [ K then all its
faces f [ K and (ii) the intersection of any two si , sj [ K is empty or a face of both si and sj.
clique complex l(G) a simplicial complex built from a graph G by promoting every k-clique c , G to a (k 2 1)-simplex
defined by the nodes of c, e.g. a 3-clique becomes a 2-simplex (a full triangle)
kth homology group Hk (K) a group describing the holes of a simplicial complex K bounded by k-dimensional boundaries, e.g. H1
describes two-dimensional cycles bounded by 1-simplices, H2 describes three-dimensional voids bounded
by 2-simplices, etc.
Hk generator g an element of the generating set of Hk, a subset of Hk such that all elements can be expressed as
combination of generators
homological scaffold H(K) a weighted graph constructed from the persistent homology generators of H1 of a simplicial complex K

between information completeness and clarity. In fact, it risks hole, we mean a hollow region bounded by the parts of that
overlooking the role that weak links might have, as shown for space. The dimension of a hole is directly related to the dimen-
example in the cases of resting-state dynamics [19,20], cognitive sion of its boundary. The boundary of a two-dimensional
control [21] and correlated network states [22]. hole is a one-dimensional loop; the three-dimensional inner
In order to overcome these limits, Rubinov & Sporn part of a doughnut, where the filling goes, is bounded by
[13,23,24] recently introduced a set of generalized network and two-dimensional surface; for dimensions higher than 2, it
community metrics for functional networks that among others becomes difficult to have a mental representation of a hole,
were used to uncover the contrasting dynamics underlying but k-dimensional holes are still bounded by (k 2 1) dimen-
recollection [25] and the physiology of functional hubs [26]. sional faces. In our work, we start with a network and from
In this paper, we present an alternative route to the analy- it construct a topological space. We now use figure 1 to
sis of brain functional networks. We focus on the combined show how we proceed and make rigorous what we mean by
structure of connections and weights as captured by the hom- boundaries and holes.
ology of the network. A summary of all the keywords and In a network like that of figure 1a, we want the ring of nodes
concepts introduced in this paper can be found in table 1. (a,b,c,d) to be a good candidate for a one-dimensional boundary,
whereas the other rings of three nodes should not constitute
interesting holes. The reason for this choice comes from the for-
2. From networks to topological spaces and malization of the notion of hole. One way to formalize this is by
opposition that is we define what we mean by a dense subnet-
homology work in order to highlight regions of reduced connectivity, i.e.
Homology is a topological invariant that characterizes a topo- holes. The most natural and conservative definition we can
logical space X by counting its holes and their dimensions. By adopt for a dense subnetwork is that of clique, a completely
connected subgraph [27]. Moreover, cliques have the crucial network structure [45] also exist. The central idea is the con- 3
property, which will be important later, of being nested, i.e. a struction of a sequence of successive approximations of the

rsif.royalsocietypublishing.org
clique of dimension k (k-clique) contains all the m-cliques original dataset seen as a topological space X. This sequence
defined by its nodes with m , k. Using this definition and filling of topological spaces X0, X1, . . . , XN ¼ X is such that
in all the maximal cliques, the network in figure 1a can be rep- Xi # Xj whenever i , j and is called the filtration. Choosing
resented as in figure 1b: 3-cliques are filled in, becoming tiles, how to construct a filtration from the data is equivalent to
and the only interesting structure left is the square (a,b,c,d). It choosing the type of goggles one wears to analyse the data.
is important at this point to note that a k-clique can be seen as In our case, we sort the edge weights in descending
a k 2 1 simplex, i.e. as the convex hull of k-points. Our represen- order and use the ranks as indices for the subspaces. More
tation of a network can thus be seen as a topological space specifically, denote by V ¼ (V, E, v ) the functional network
formed by a finite set of simplices that by construction satisfy with vertices V, edges E and weights v  : E ! R. We then con-
the condition that defines the type of topological spaces called sider the family of binary graphs Gv ¼ (V, Ev), where an edge

J. R. Soc. Interface 11: 20140873

abstract simplicial complexes [28]: each element of the space is e [ E is also included in Gv if its weight ve is larger than v
a simplex, and each of its faces (or subset in the case of (e [ Ev , v  (e)  v).
cliques) is also a simplex. To each of the Gv, we associate its clique, or flag complex Kv,
This condition is satisfied, because each clique is a sim- that is the simplicial complex that contains the k-simplex [n0,
plex, and subsets of cliques are cliques themselves, and the n1, n2, . . . nk2 1] whenever the nodes n0, n1, n2, . . . nk21 define
intersection of two cliques is still a clique. a clique in Gv [27]. As subsets of cliques and intersections of
The situation with weighted networks becomes more cliques are cliques themselves, as we pointed out in §2, our
complicated. In the context of a weighted network, the clique complex is thus a particular case of a simplicial
holes can be thought of as representing regions of reduced complex.
connectivity with respect to the surrounding structure. The family of complexes fKvg defines a filtration, because
Consider, for example, the case depicted in figure 2a: the we have Kv # Kv0 for v . v0 . At each step, the simplices in Kv
network is almost the same as figure 1 with the two excep- inherit their configuration from the underlying network
tions that it now has weighted edges and has an additional structure and, because the filtration swipes across all weight
very weak edge between nodes a and c. The edges in the scales in descending order, the holes among these units con-
cycle [a,b,c,d] are all much stronger than the link (a,c) that stitute mesoscopic regions of reduced functional connectivity.
closes the hole by making (a,b,d) and (b,c,d) cliques and there- Moreover, this approach also highlights how network
fore fills them. The loop (e,f,g,h,i) has a similar situation, but properties evolve along the filtration, providing insights
the difference in edge weights between the links along the about where and when lower connectivity regions emerge.
cycle and those crossing, is not as large as in the previous This information is available, because it is possible to keep
case. It would be therefore useful to be able to generalize track of each k-dimensional cycle in the homology group Hk.
the approach exposed earlier for binary networks to the case A generator uniquely identifies a hole by its constituting
of weighted networks in such a way as to be able to measure elements at each step of the filtration process. The importance
the difference between the two cases (a,b,c,d) and (e,f,g,h,i). As of a hole is encoded in the form of ‘time-stamps’ recording its
shown by figure 2b, this problem can be intuitively thought of birth bg and death dg along the filtration fKvg [31]. These two
as a stratigraphy in the link-weight fabric of the network, time-stamps can be combined to define the persistence pg ¼
where the aim is to detect the holes, measure their depth dg 2 bg of a hole, which gives a notion of its importance in
and when they appear as we scan across the weights’ range. terms of it lifespan. Continuing the analogy with stratigraphy,
From figure 2b, it becomes clear that the added value of bg and dg correspond, respectively, to the top and the bottom
this method over conventional network techniques lies in its of a hole and pg would be its depth. As we said above, a genera-
capability to describe mesoscopic patterns that coexist over tor gki , or hole, of the kth homology group Hk is identified by its
different intensity scales, and hence to complement the infor- birth and death along the filtration. Therefore, gki is described by
mation about the community structure of brain functional the point (bg , dg ) [ R2 . A standard way to summarize the infor-
networks. A way to quantify the relevance of holes is given mation about the whole kth persistent homology group is then
by persistent homology. We describe it and its application to the to consider the diagram obtained plotting the points corre-
case of weighted networks in full detail in §3. sponding to the set of generators. The (multi)set f(bg,dgg is
called the persistence diagram of Hk. In figure 2c, we show the
persistence diagram for the network shown in figure 2a for
H1. Axes are labelled by weights in decreasing order. It is easy
3. A persistent homology of weighted networks to check that the coordinates correspond exactly to the appear-
The method that we adopt was introduced in references ance and disappearance of generators. The green vertical bars
[29,30] and relies on an extension of the metrical persistent highlight the persistence of a generator along the filtration.
homology theory originally introduced by references [31,32]. The further a point is from the diagonal (vertically), the more
Technical details about the theory of persistent homology persistent the generator is. In §4, we introduce two objects, the
and how the computation is performed can be found persistence and the frequency homological scaffolds, designed
in the works of Carlsson, Zomorodian and Edelsbrunner to summarize the topological information about the system.
[28,31–35]. Persistent homology is a recent technique in com-
putational topology developed for shape recognition and
the analysis of high dimensional datasets [36,37]. It has been
used in very diverse fields, ranging from biology [38,39] 4. Homological scaffolds
and sensor network coverage [40] to cosmology [41]. Similar Once one has calculated the generators {gki }i of the kth persist-
approaches to brain data [42,43], collaboration data [44] and ent homology group Hk, the corresponding persistence
 (a) k 4
8
d

rsif.royalsocietypublishing.org
8
j 10 a

8
10 1
0 c
b 1

10
10
i
6
e
5

4
h 4 f
6

6
8 g
l

J. R. Soc. Interface 11: 20140873

(b)
10

appearance of
(a,b,c,d) hole
8
appearance of
(e,f,g,h,i) hole
6
weights

4
closure of
(e, f,g,h,i) hole
2 closure of
(a,b,c,d) hole

(c)
0
(a,b,c,d)
1
2
3
(e,f,g,h,i)
death weight

4
5
6
7
8

9 8 7 6 5 4 3 2 1 0
birth weight
Figure 2. Panels (a – c) display a weighted network (a), its intuitive representation in terms of a stratigraphy in the weight structure according the weight filtration
described in the main text (b) and the persistence diagram for H1 associated with the network shown (c). By promoting cliques to simplices, we identify network
connectivity with relations between the vertices defining the simplicial complex. By producing a sequence of networks through the filtration, we can study the
emergence and relative significance of specific features along the filtration. In this example, the hole defined by (a,b,c,d ) has a longer persistence (vertical solid
green bars) implying that the boundary of the cycle are much heavier than the internal links that eventually close it. The other hole instead has a much shorter
persistence, surviving only for one step and is therefore considered less important in the description of the network homological properties. Note that the births and
deaths are defined along the sequence of descending edge weights in the network, not in time. (Online version in colour.)

diagram contains a wealth of information that can be used, at least for low dimensions k. Here, we present a scheme to
for example, to highlight differences between two datasets. obtain such a description by using the information associated
It would be instructive to obtain a synthetic description of with the generators during the filtration process. As
the uncovered topological features in order to interpret the each generator, gki is associated with a whole equivalence
observed differences in terms of the microscopic components, class, rather than to a single chain of simplices, we need to
choose a representative for each class, we use the representa- differences detected by the persistent homology actually 5
tive that is returned by the javaplex implementation [46] of the mean in terms of the system under study.

rsif.royalsocietypublishing.org
persistent homology algorithm [47]. For the sake of simplicity
in the following, we use the same symbol gki to refer to a
generator and its representative cycle.
We exploit this to define two new objects, the persistence and 5. Results from fMRI networks
p f
the frequency homological scaffolds HG and HG of a graph G. The We start from the processed fMRI time series (see Methods for
persistence homological scaffold is the network composed of all details). The linear correlations between regional time series
the cycle paths corresponding to generators weighted by their were calculated after covarying out the variance owing to all
persistence. If an edge e belongs to multiple cycles g0,g1, . . . ,gs, other regions and the residual motion variance represented
its weight is defined as the sum of the generators’ persistence: by the 24 rigid motion parameters obtained from the pre-
processing, yielding a partial-correlation matrix xa for each

J. R. Soc. Interface 11: 20140873

X
vpe ¼ pgi : (4:1) subject. The matrices xa were then analysed with the algorithm
gi je[gi described in the previous sections. We calculated the generators
f g1i of the first homological group H1 along the filtration. As
Similarly, we define the frequency homological scaffold HG as
mentioned before, each of these generators identifies a lack of
the network composed of all the cycle paths corresponding
mesoscopic connectivity in the form of a one-dimensional
to generators, where this time, an edge e is weighted by the
cycle and can be represented in a persistence diagram. We aggre-
number of different cycles it belongs to
gate together the persistence diagrams of subjects belonging to
X each group and compute an associated persistence probability
vfe ¼ 1e[gi , (4:2)
gi
density (figure 3). These probability density functions constitute
the statistical signature of the groups’ H1 features.
where 1e[gi is the indicator function for the set of edges com- We find that, although the number of cycles in the groups
posing gi. By definition, the two scaffolds have the same edge are comparable, the two probability densities strongly differ
set, although differently weighted. (Kolmogorov–Smirnov statistics: 0.22, p-value less than 10210).
The construction of these two scaffolds therefore high- The placebo group displays generators appearing and per-
lights the role of links which are part of many and/or long sisting over a limited interval of the filtration. On the contrary,
persistence cycles, isolating the different roles of edges most of the generators for the psilocybin group are situated in
within the functional connectivity network. The persistence a well-defined peak at small birth indices, indicating a shorter
scaffolds encodes the overall persistence of a link through average cycle persistence. However, the psilocybin distribution
the filtration process: the weight in the persistence scaffold is also endowed with a longer tail implying the existence of
of a link belonging to a certain set of generators is equal to a few cycles that are longer-lived compared with the placebo
the sum of the persistence of those cycles. The frequency scaf- condition and that influences the weight distribution of the psi-
fold instead highlights the number of cycles to which a link locybin persistence scaffold. The difference in behaviour of the
belongs, thus giving another measure of the importance of two groups is made explicit when looking at the probability dis-
that edge during the filtration. The combined information tribution functions for the persistence and the birth of
given by the two scaffolds then enables us to decipher the generators (figure 4), which are both found to be significantly
nature of the role different links have regarding the homolo- different (Kolmogorov–Smirnov statistics: 0.13, p-value ,
gical properties of the system. A large total persistence for a 10230 for persistence and Kolmogorov–Smirnov statistics:
link in the persistence scaffold implies that the local structure 0.14, p-value , 10235 for births). In order to better interpret
around that link is very weak when compared with the and understand the differences between the two groups,
f
weight of the link, highlighting the link as a locally strong we use the two secondary networks described in §4, HPla and
p f p
bridge. We remark that the definition of scaffolds we gave HPla for the placebo group and HPsi and HPsi for the psilocybin
depends on the choice of a specific basis of the homology group. The weight of the edges in these secondary networks is
group, and the choice of a consistent basis is an open problem proportional to the total number of cycles an edge is part of, and
in itself, therefore the scaffolds are not topological invariants. the total persistence of those cycles, respectively. They comp-
Moreover, it is possible for an edge to be added to a cycle lement the information given by the persistence density
shortly after the cycle’s birth in such a way that it creates a distribution, where the focus is on the entire cycle’s behaviour,
triangle with the two edges composing the cycle. In this with information on single links. In fact, individual edges
way, the new edge would be part of the shortest cycle, but belonging to many and long persistence cycles represent func-
the scaffold persistence value would be misattributed to the tionally stable ‘hub’ links. As with the persistence density
two other edges. This can be checked, for example, by moni- distribution, the scaffolds are obtained at a group level by
toring the clustering coefficient of the cycle’s subgraph as aggregating the information about all subjects in each group.
edges are added to it. We checked for this effect and found These networks are slightly sparser than the original complete
that in over 80% of the cases the edges do not create triangles xa networks
that would imply the error, but instead new cycles are cre- p
f,p
2m(Hpla )
ated, whose contribution to the scaffold is then accounted r(Hpla ) ¼ ¼ 0:92 (5:1)
n(n  1)
for by the new cycle. Finally, we note also that, when a
new triangle inside the cycle is created, the two choices of and
generator differ for a path through a third strongly connected p
f,p 2m(HPsi )
node, owing to the properties of boundary operators. Despite r(HPsi ) ¼ ¼ 0:91 (5:2)
n(n  1)
this ambiguity, we show in the following that they can be
useful to gain an understanding of what the topological and have comparable densities. A first difference between the
 (a) (b) 6
1.0 1.0

rsif.royalsocietypublishing.org
0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

death
death

0.5 0.5

0.4 0.4

J. R. Soc. Interface 11: 20140873

0.3 0.3

0.2 0.2

0.1 0.1

0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.00
0

0
7

5
1.

0.

0.

0.

0.

0.

0.
0.

0.

0.

birth birth

–10.0 –9.5 –9.0 –8.5 –8.0 –7.5 –7.0 –6.5 –10.0 –9.6 –9.2 –8.8 –8.4 –8.0 –7.6 –7.2 –6.8
Figure 3. Probability densities for the H1 generators. Panel (a) reports the (log-)probability density for the placebo group, whereas panel (b) refers to the psilocybin
group. The placebo displays a uniform broad distribution of values for the births – deaths of H1 generators, whereas the plot for the psilocybin condition is very
peaked at small values with a fatter tail. These heterogeneities are evident also in the persistence distribution and find explanation in the different functional
integration schemes in placebo and drugged brains. (Online version in colour.)

(a) (b)
0.05 0.1
placebo
psilo
0.04
0.01

0.03
P(b)
P(p)

0.001
0.02

0.0001
0.01

10–5
0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0
persistence p birth b
Figure 4. Comparison of persistence p and birth b distributions. Panel (a) reports the H1 generators’ persistence distributions for the placebo group (blue line) and
psilocybin group (red line). Panel (b) reports the distributions of births with the same colour scheme. It is very easy to see that the generators in the psilocybin
condition have persistence peaked at shorter values and a wider range of birth times when compared with the placebo condition. (Online version in colour.)

two groups becomes evident when we look at the distributions persistence scaffolds of the same group. Figure 5b is a scatter
for the edge weights (figure 5a). In particular, the weights of plot of between the weights of edges from both scaffolds for
p
Hpla display a cut-off for large weights, whereas the weights the two groups. The placebo group has a linear relationship
p
of Hpsi have a broader tail (Kolmogorov–Smirnov statistics: between the two quantities meaning that edges that are per-
0.06, p-value , 10220; figure 5a). Interestingly, the frequency sistent also belongs to many cycles (R 2 ¼ 0.95, slope ¼ 0.23).
scaffold weights probability density functions cannot be distin- Although the linear relationship is still a reasonable fit for
guished from each other figure 5a (inset) (Kolmogorov– the psilocybin group (R 2 ¼ 0.9, slope ¼ 0.3), the data in this
Smirnov statistics: 0.008, p-value ¼ 0.72). Taken together, case display a larger dispersion. In particular, it shows that
f,p
these two results imply that while edges statistically belong to edges in Hpsi can be much more persistent/longer-lived
f,p
the same number of cycles, in the psilocybin scaffold, there than in Hpla but still appear in the same number of cycles,
exist very strong, persistent links. i.e. the frequency of a link is not predictive of its persistence
The difference between the two sets of homological scaf- or simply put: some connections are much more persistent in
folds for the two groups becomes even more evident when the psychedelic state. Moreover, the slopes of linear fits of the
one compares the weights between the frequency and two clouds are statistically different ( p-value , 10220, npla ¼
 (a) (b) 7
1 350

rsif.royalsocietypublishing.org
p psi
pla 300
p pla
10–1 psi

link persistence wep

250

200
P(wep )

10–2
1.0 150
0.1
P(wef )

0.01 100
10–3
0.001

J. R. Soc. Interface 11: 20140873

0.0001
50
10.0 100.0 1000.0
link frequency w ef
10–4
10 102 103 0 200 400 600 800 1000 1200
link persistence wep link frequency wef

Figure 5. Statistical features of group homological scaffolds. Panel (a) reports the (log-binned) probability distributions for the edge weights in the persistence
homological scaffolds (main plot) and the frequency homological scaffolds (inset). While the weights in the frequency scaffold are not significantly different, the
weight distributions for the persistence scaffold display clearly a broader tail. Panel (b) shows instead the scatter plot of the edge frequency versus total persistence.
In both cases, there is a clear linear relationship between the two, with a large slope in the psilocybin case. Moreover, the psilocybin scaffold has a larger spread in
the frequency and total persistence of individual edges, hinting to a different local functional structure within the functional network of the drugged brains. (Online
version in colour.)

13 200 and npsi ¼ 13 275 [48]) pointing to a starkly different implies that the brain does not simply become a random
local functional structure in the two conditions. system after psilocybin injection, but instead retains some organ-
The results from the persistent homology analysis and the izational features, albeit different from the normal state, as
insights provided by the homological scaffolds imply that suggested by the first part of the analysis. Further work is
although the mesoscopic structures, i.e. cycles, in the psilocy- required to identify the exact functional significance of these
bin condition are less stable than in the placebo group, their edges. Nonetheless, it is interesting to look at the community
constituent edges are more stable. structure of the persistence homological scaffolds in figure 6.
The two pictures are simplified cartoons of the placebo (figure
6a) and psilocybin (figure 6b) scaffolds. In figure 6a,b, the
nodes are organized and coloured according to their community
6. Discussion membership in the placebo scaffold (obtained with the Louvain
In this paper, we first described a variation of persistent hom- algorithm for maximal modularity and resolution 1 [50]). This is
ology that allows us to deal with weighted and signed done in order to highlight the striking difference in connectivity
networks. We then introduced two new objects, the homolo- structure in the two cases. When considering the edges in the tail
gical scaffolds, to go beyond the picture given by persistent of the distribution, weight greater than or equal to 80, in figure
homology to represent and summarize information about 5a, only 29 of the 374 edges present in the truncated psilocybin
individual links. The homological scaffolds represent a new scaffold are shared with the truncated placebo scaffold (165
measure of topological importance of edges in the original edges). Of these 374 edges, 217 are between placebo commu-
system in terms of how frequently they are part of the genera- nities and are observed to mostly connect cortical regions. This
tors of the persistent homology groups and how persistent supports our idea that psilocybin disrupts the normal organiz-
are the generators to which they belong to. We applied this ation of the brain with the emergence of strong, topologically
method to an fMRI dataset comprising a group of subjects long-range functional connections that are not present in a
injected with a placebo and another injected with psilocybin. normal state.
By focusing on the second homology group H1, we found The two key results of the analysis of the homological scaf-
that the stability of mesoscopic association cycles is reduced by folds can therefore be summarized as follows (i) there is an
the action of psilocybin, as shown by the difference in the increased integration between cortical regions in the psilocybin
probability density function of the generators of H1 (figure 3). state and (ii) this integration is supported by a persistent scaffold
It is here that the importance of the insight given by the of a set of edges that support cross modular connectivity prob-
homological scaffolds in the persistent homology procedure ably as a result of the stimulation of the 5HT2A receptors in
becomes apparent. A simple reading of this result would be the cortex [51].
that the effect of psilocybin is to relax the constraints on brain We can speculate on the implications of such an organiz-
function, ascribing cognition a more flexible quality, but when ation. One possible by-product of this greater communication
looking at the edge level, the picture becomes more complex. across the whole brain is the phenomenon of synaesthesia
The analysis of the homological scaffolds reveals the existence which is often reported in conjunction with the psychedelic
of a set of edges that are predominant in terms of their persist- state. Synaesthesia is described as an inducer-concurrent
ence although they are statistically part of the same number of pairing, where the inducer could be a grapheme or a visual
cycles in the two conditions (figure 5). In other words, these stimulus that generates a secondary sensory output—like a
functional connections support cycles that are especially stable colour for example. Drug-induced synaesthesia often leads
and are only present in the psychedelic state. This further to chain of associations, pointing to dynamic causes rather
 (a) (b) 8

rsif.royalsocietypublishing.org
J. R. Soc. Interface 11: 20140873
Figure 6. Simplified visualization of the persistence homological scaffolds. The persistence homological scaffolds Hplap (a) and Hpsip (b) are shown for comparison.
For ease of visualization, only the links heavier than 80 (the weight at which the distributions in figure 5a bifurcate) are shown. This value is slightly smaller than
the bifurcation point of the weights distributions in figure 5a. In both networks, colours represent communities obtained by modularity [49] optimization on the
placebo persistence scaffold using the Louvain method [50] and are used to show the departure of the psilocybin connectivity structure from the placebo baseline.
The width of the links is proportional to their weight and the size of the nodes is proportional to their strength. Note that the proportion of heavy links between
communities is much higher (and very different) in the psilocybin group, suggesting greater integration. A labelled version of the two scaffolds is available as GEXF
graph files as the electronic supplementary material. (Online version in colour.)

than fixed structural ones as may be the case for acquired functional data to the middle volume of the acquisition using
synaesthesia [52]. Broadly consistent with this, it has been the FMRIB linear registration motion correction tool, generating
reported that subjects under the influence of psilocybin a six-dimension parameter time course [55]. Recent work demon-
have objectively worse colour perception performance strates that the six parameter motion model is insufficient to
correct for motion-induced artefact within functional data,
despite subjectively intensified colour experience [53].
instead a Volterra expansion of these parameters to form a 24
To summarize, we presented a new method to analyse
parameter model is favoured as a trade-off between artefact cor-
fully connected, weighted and signed networks and applied
rection and lost degrees of freedom as a result of regressing
it to a unique fMRI dataset of subjects under the influence motion away from functional time courses [56]. fMRI data
of mushrooms. We find that the psychedelic state is associ- were pre-processed according to standard protocols using a
ated with a less constrained and more intercommunicative high-pass filter with a cut-off of 300 s.
mode of brain function, which is consistent with descriptions Structural MRI images were segmented into n ¼ 194 cortical
of the nature of consciousness in the psychedelic state. and subcortical regions, including white matter cerebrospinal
fluid (CSF) compartments, using FREESURFER (http://surfer.nmr.
mgh.harvard.edu/), according to the Destrieux anatomical atlas
[57]. In order to extract mean-functional time courses from
7. Methods the BOLD fMRI, segmented T1 images were registered to the
middle volume of the motion-corrected fMRI data, using bound-
7.1. Dataset ary-based registration [58], once in functional space mean
A pharmacological MRI dataset of 15 healthy controls was used
time-courses were extracted for each of the n ¼ 194 regions in
for a proof-of-principle test of the methodology [54]. Each subject
native fMRI space.
was scanned on two separate occasions, 14 days apart. Each scan
consisted of a structural MRI image (T1-weighted), followed by a
12 min eyes-close resting-state blood oxygen-level-dependent 7.1.3. Functional connectivity
(BOLD) fMRI scan which lasted for 12 min. Placebo (10 ml For each of the 194 regions, alongside the 24 parameter motion
saline, intravenous injection) was given on one occasion and psi- model time courses, partial correlations were calculated between
locybin (2 mg dissolved in 10 ml saline) on the other. Injections all couples of time courses (i,j), non-neural time courses (CSF,
were given manually by a study doctor situated within the scan- white matter and motion) were discarded from the resulting
ning suite. Injections began exactly 6 min after the start of the functional connectivity matrices, resulting in a 169 region corti-
12-min scans, and continued for 60 s. cal/subcortical functional connectivity corrected for motion
and additional non-neural signals (white matter/CSF).
7.1.1. Scanning parameters
The BOLD fMRI data were acquired using standard gradient-echo
EPI sequences, reported in detail in reference [54]. The volume
7.2. Persistent homology computation
For each subject in the two groups, we have a set of persistence
repetition time was 3000 ms, resulting in a total of 240 volumes
diagrams relative to the persistent homology groups Hn. In this
acquired during each 12 min resting-state scan (120 pre- and 120
paper, we use the H1 persistence diagrams of each group to
post-injection of placebo/psilocybin).
construct the corresponding persistence probability densities
for H1 cycles.
7.1.2. Image pre-processing Filtrations were obtained from the raw partial-correlation
fMRI images were corrected for subject motion within individual matrices through the PYTHON package Holes and fed to javaplex
resting-state acquisitions, by registering all volumes of the [46] via a Jython subroutine in order to extract the persistence
intervals and the representative cycles. The details of the the European Commission under Contract IST-318121. I.D. P.E. and 9
implementation can be found in reference [30], and the software F.T. are supported by a PET methodology programme grant from
is available at Holes [59]. the Medical Research Council UK (ref no. G1100809/1). The authors

rsif.royalsocietypublishing.org
acknowledge support of Amanda Feilding and the Beckley
Funding statement. G.P. and F.V. are supported by the TOPDRIM project Foundation and the anonymous referees for their critical and
supported by the Future and Emerging Technologies programme of constructive contribution to this paper.

References
1. Baronchelli A, Ferrer-i-Cancho R, Pastor-Satorras R, interpretations. Neuroimage 52, 1059–1069. 26. Liang X, Zou Q, He Y, Yang Y. 2013 Coupling of
Chater N, Christiansen MH. 2013 Networks in (doi:10.1016/j.neuroimage.2009.10.003) functional connectivity and regional cerebral blood

J. R. Soc. Interface 11: 20140873

cognitive science. Trends Cogn. Sci. 17, 348 –360. 14. Zuo XN, Ehmke R, Mennes M, Imperati D. 2012 flow reveals a physiological basis for network hubs
(doi:10.1016/j.tics.2013.04.010) Network centrality in the human functional of the human brain. Proc. Natl Acad. Sci. USA 110,
2. Bullmore E, Sporns O. 2009 Complex brain connectome. Cereb. Cortex 22, 1862–1875. 1929– 1934. (doi:10.1073/pnas.1214900110)
networks: graph theoretical analysis of structural (doi:10.1093/cercor/bhr269) 27. Evans TS. 2010 Clique graphs and overlapping
and functional systems. Nat. Rev. Neurosci. 10, 15. Qian S, Sun G, Jiang Q, Liu K, Li B, Li M, Yang X, communities. JSTAT 2010, P12037. (doi:10.1088/
186–198. (doi:10.1038/nrn2618) Yang Z, Zhao L. 2013 Altered topological patterns of 1742-5468/2010/12/P12037)
3. Meunier D, Lambiotte R, Bullmore E. 2010 Modular large-scale brain functional networks during passive 28. Edelsbrunner H, Harer J. 2010 Computational
and hierarchically modular organization of brain hyperthermia. Brain Cogn. 83, 121–131. (doi:10. topology: an introduction. Providence, RI: American
networks. Front. Neurosci. 4, 200. (doi:10.3389/ 1016/j.bandc.2013.07.013) Mathematical Society.
fnins.2010.00200) 16. Sepulcre J, Liu H, Talukdar T, Martincorena I. 2010 29. Petri G, Scolamiero M, Donato I, Vaccarino F. 2013
4. Pandit AS, Expert P, Lambiotte R, Bonnelle V, Leech The organization of local and distant functional Networks and cycles: a persistent homology approach
R, Turkheimer FE, Sharp DJ. 2013 Traumatic brain connectivity in the human brain. PLoS Comput. Biol. to complex networks. In Proc. the European Conf. on
injury impairs small-world topology. Neurology 80, 6, e1000808. (doi:10.1371/journal.pcbi.1000808) Complex Systems 2012 (eds T Gilbert, M Kirkilionis,
1826–1833. (doi:10.1212/WNL.0b013e3182929f38) 17. Ginestet CE, Nichols TE, Bullmore ET, Simmons A. G Nicolis), pp. 93–99. Springer Proceedings in
5. Lord LD et al. 2012 Functional brain networks 2011 Brain network analysis: separating cost from Complexity. Berlin, Germany: Springer. (doi:10.1007/
before the onset of psychosis: a prospective fMRI topology using cost-integration. PLoS ONE 6, 978-3-319-00395-5)
study with graph theoretical analysis. Neuroimage: e21570. (doi:10.1371/journal.pone.0021570) 30. Petri G, Scolamiero M, Donato I, Vaccarino F. 2013
Clin. 1, 91 –98. (doi:10.1016/j.nicl.2012.09.008) 18. Power JD, Cohen AL, Nelson SM, Wig GS, Barnes KA. Topological strata of weighted complex networks.
6. Lord LD, Allen P, Expert P, Howes O, Lambiotte R, 2011 Functional network organization of the human PLoS ONE 8, e66506. (doi:10.1371/journal.pone.
McGuire P, Bose SK, Hyde S, Turkheimer FE. 2011 brain. Neuron 72, 665– 678. (doi:10.1016/j.neuron. 0066506)
Characterization of the anterior cingulate’s role in 2011.09.006) 31. Edelsbrunner H, Letscher D, Zomorodian A. 2002
the at-risk mental state using graph theory. 19. Schwarz AJ, McGonigle J. 2011 Negative edges and Topological persistence and simplification. Discrete
Neuroimage 56, 1531 –1539. (doi:10.1016/j. soft thresholding in complex network analysis of Comput. Geom. 28, 511–533. (doi:10.1007/s00454-
neuroimage.2011.02.012) resting state functional connectivity data. 002-2885-2)
7. Zalesky A, Fornito A, Bullmore E. 2012 On the use J. Neuroimage 55, 1132–1146. (doi:10.1016/j. 32. Carlsson G, Zomorodian A. 2005 Computing
of correlation as a measure of network connectivity. neuroimage.2010.12.047) persistent homology. Discrete Comput. Geom. 33,
Neuroimage 60, 2096 –2106. (doi:10.1016/j. 20. Bassett DS, Nelson BG, Mueller BA, Camchong J, 249–274. (doi:10.1007/s00454-004-1146-y)
neuroimage.2012.02.001) Lim KO. 2012 Altered resting state complexity in 33. Steiner DC, Edelsbrunner H, Harer J. 2007 Stability
8. Achard S, Salvador R, Whitcher B, Suckling J, schizophrenia. Neuroimage 59, 2196– 2207. of persistence diagrams. Discrete Comput. Geom. 37,
Bullmore E. 2006 A resilient, low-frequency, small- (doi:10.1016/j.neuroimage.2011.10.002) 103–120. (doi:10.1007/s00454-006-1276-5)
world human brain functional network with highly 21. Cole MW, Yarkoni T, Repovs G, Anticevic A, Braver TS. 34. Carlsson G. 2009 Topology and data. Bull. Am. Math.
connected association cortical hubs. J. Neurosci. 21, 2012 Global connectivity of prefrontal cortex Soc. 46 255–308. (doi:10.1090/S0273-0979-09-
63 –72. (doi:10.1523/JNEUROSCI.3874-05.2006) predicts cognitive control and intelligence. J. Neurosci. 01249-X)
9. Newman MEJ. 2001 Scientific collaboration 32, 8988–8999. (doi:10.1523/JNEUROSCI.0536- 35. Ghrist R. 2008 Barcodes: the persistent topology of
networks. II. Shortest paths, weighted networks, 12.2012) data. Bull. Am. Math. Soc. 45, 61 –75. (doi:10.1090/
and centrality. Phys. Rev. E 64, 016132. (doi:10. 22. Schneidman E, Berry MJ, Segev R, Bialek W. 2006 S0273-0979-07-01191-3)
1103/PhysRevE.64.016132) Weak pairwise correlations imply strongly correlated 36. Carlsson G, Ishkhanov T, Silva V, Zomorodian A.
10. Opsahl T, Agneessens F, Skvoretz J. 2010 Node network states in a neural population. Nature 440, 2008 On the local behaviour of spaces of natural
centrality in weighted networks: generalizing 1007 –1012. (doi:10.1038/nature04701) images. Int. J. Comput. Vision 76, 1–12. (doi:10.
degree and shortest paths. Soc. Netw. 32, 245 – 23. Rubinov M, Sporns O. 2011 Weight-conserving 1007/s11263-007-0056-x)
251. (doi:10.1016/j.socnet.2010.03.006) characterization of complex functional brain 37. Lum PY et al. 2012 Extracting insights from the
11. Opsahl T, Colizza V, Panzarasa P, Ramasco JJ. 2008 networks. Neuroimage 56, 2068–2079. (doi:10. shape of complex data using topology. Sci. Rep. 3,
Prominence and control: the weighted rich-club 1016/j.neuroimage.2011.03.069) 1236. (doi:10.1038/srep01236)
effect. Phys. Rev. Lett. 101, 168702. (doi:10.1103/ 24. Sporns O. 2013 Structure and function of complex 38. Chan JM, Carlsson G, Rabadan R. 2013 Topology of
PhysRevLett.101.168702) brain networks. Dialogues Clin. Neurosci. 15, viral evolution. Proc. Natl Acad. Sci. USA 110,
12. Colizza V, Flammini A, Serrano MA, Vespignani A. 247 –262. 18 566 –18 571. (doi:10.1073/pnas.1313480110)
2006 Detecting rich-club ordering in complex 25. Fornito A, Harrison BJ, Zalesky A, Simons JS. 2012 39. Nicolau M, Levine A, Carlsson G. 2011 Topology
networks. Nat. Phys. 2, 110–115. (doi:10.1038/ Competitive and cooperative dynamics of large-scale based data analysis identifies a subgroup of breast
nphys209) brain functional networks supporting recollection. cancers with a unique mutational profile and
13. Rubinov M, Sporns O. 2010 Complex network Proc. Natl Acad. Sci. USA 109, 12 788–12 793. excellent survival. Proc. Natl Acad. Sci. USA 108,
measures of brain connectivity: uses and (doi:10.1073/pnas.1204185109) 7265– 7270. (doi:10.1073/pnas.1102826108)
40. De Silva V, Ghrist R. 2007 Coverage in sensor topological invariants. See http://git. 54. Carhart-Harris RL et al. 2012 Neural correlates of 10
networks via persistent homology. Algebr. Geom. appliedtopology.org/javaplex/. the psychedelic state as determined by fMRI

rsif.royalsocietypublishing.org
Topol. 7, 339– 358. (doi:10.2140/agt.2007.7.339) 47. Skraba P, Vejdemo-Johansson M. 2013 Persistence studies with psilocybin. Proc. Natl Acad. Sci.
41. van de Weygaert R et al. 2011 Alpha, betti and the modules: algebra and algorithms. (http://arxiv.org/ USA 109, 2138–2143. (doi:10.1073/pnas.
megaparsec universe: on the topology of the cosmic abs/1302.2015) 1119598109)
web. Trans. Comput. Sci. XIV, 60 –101. Special issue 48. Cohen J, Cohen P. 1983 Applied multiple regression/ 55. Smith SM et al. 2004 Advances in functional and
on Voronoi diagrams and Delaunay triangulation. correlation analysis for the behavioral sciences. structural MR image analysis and implementation
(http://arxiv.org/abs/1306.3640) Hillsdale, NJ: Lawrence Erlbaum Associates. as FSL. Neuroimage 23, 208–219. (doi:10.1016/j.
42. Lee H, Chung MK, Kang H, Kim B-N, Lee DS. 2011 49. Newman MEJ. 2006 Modularity and neuroimage.2004.07.051)
Discriminative persistent homology of brain community structure in networks. Proc. Natl Acad. 56. Satterthwaite TD et al. 2013 An improved
networks. In IEEE Int. Symp. on Biomedical Imaging: Sci. USA 103, 8577–8582. (doi:10.1073/pnas. framework for confound regression and filtering for
From Nano to Macro, 30 March 2011– 2 April 2011, 0601602103) control of motion artifact in the preprocessing of

J. R. Soc. Interface 11: 20140873

pp. 841 –844. Piscataway, NJ: IEEE. (doi:10.1109/ 50. Blondel VD, Guillaume JL, Lambiotte R, Lefebvre E. resting-state functional connectivity data.
ISBI.2011.5872535) 2008 Fast unfolding of communities in large Neuroimage 64, 240– 256. (doi:10.1016/j.
43. Lee H, Kang H, Chung MK, Kim BN, Lee DS. 2012 networks. J. Stat. Mech. Theory Exp. 2008, P10008. neuroimage.2012.08.052)
Persistent brain network homology from the (doi:10.1088/1742-5468/2008/10/P10008) 57. Destrieux C, Fischl B, Dale A, Halgren E. 2010
perspective of dendrogram. IEEE Trans. Med. Imaging 51. Eastwood SL, Burnet PW, Gittins R, Baker K, Harrison PJ. Automatic parcellation of human cortical gyri and
31, 2267–2277. (doi:10.1109/TMI.2012.2219590) 2001 Expression of serotonin 5-HT(2A) receptors in the sulci using standard anatomical nomenclature.
44. Carstens CJ, Horadam KJ. 2013 Persistent homology human cerebellum and alterations in schizophrenia. Neuroimage 15, 1–15. (doi:10.1016/j.neuroimage.
of collaboration networks. Math. Probl. Eng. 2013 Synapse 42, 104–114. (doi:10.1002/syn.1106) 2010.06.010)
1– 7. (doi:10.1155/2013/815035) 52. Sinke C, Halpern JH, Zedler M, Neufeld J, Emrich HM, 58. Greve DN, Fischl B. 2009 Accurate and robust brain
45. Danijela H, Maletić S, Rajković M. 2009 Persistent Passie T. 2012 Genuine and drug-induced synesthesia: image alignment using boundary-based registration.
homology of complex networks. J. Stat. Mech. a comparison. Conscious. Cogn. 21, 1419–1434. Neuroimage 48, 63–72. (doi:10.1016/j.neuroimage.
2009, P03034. (doi:10.1088/1742-5468/2009/03/ (doi:10.1016/j.concog.2012.03.009) 2009.06.060)
P03034) 53. Hartman AM, Holister LE. 1963 Effect of mescaline, 59. Computational Topology workgroup, Stanford
46. Computational Topology workgroup, Stanford lysergic acid diethylamide and psylocibin on colour University. 2013 Holes: Java library for computing
University. 2014 JavaPlex: Java library for perception. Psychomarmacolgia 4, 441 –451. persistent homology and other topological invariants.
computing persistent homology and other (doi:10.1007/BF00403349) See https://code.google.com/p/javaplex/.