Abnormal topological organization in white matter structural networks in survivors of acute lymphoblastic leukaemia with chemotherapy treatment

Previous diffusion tensor imaging (DTI) studies have detected white matter (WM) integrity abnormalities in some specific fibre bundles in acute lymphoblastic leukaemia (ALL) patients with chemotherapy. However, little is known about the changes in the topological organization of the WM structural network in ALL patients with chemotherapy. In the present study, we acquired DTI datasets from 28 ALL patients (mean age: 40.71 ± 8.58 years, years since diagnosis: 7–38) with chemotherapy and 20 matched healthy controls (mean age: 42.95 ± 6.39 years) and performed WM network analysis using a deterministic fibre-tracking approach. Graph theoretical analysis was used to compare the topological parameters of the WM networks between the two groups. Both ALL patients with chemotherapy and healthy controls had small-worldness in their WM networks. ALL patients showed significantly reduced global network efficiency, as indicated by the abnormally decreased clustering coefficient Cp and the normalized clustering coefficient γ and increased shortest path length Lp compared with healthy controls. Moreover, hubs were located more in parietal regions of healthy controls and in temporal regions in the ALL patients. We revealed the abnormal topological organization of the WM networks of ALL patients with chemotherapy, which may improve our understanding of the neural mechanism of chemotherapy in ALL from a WM topological organization level.


INTRODUCTION
Acute lymphoblastic leukaemia (ALL) is the most common cancer among children, accounting for 74% of all leukaemias and 18% of all cancers [1,2].ALL survivors suffer from long-term neurocognitive impairment and poor health-related quality of life (QoL) [3].The historical treatment of central nervous system (CNS)-directed therapy has resulted in a 5-year event-free survival rate of approximately 80% [4].CNSdirected therapies mainly consist of cranial irradiation, chemotherapy, and combinations of these two methods.Among these CNS-directed therapies, cranial irradiation has been implicated as the cause of neurocognitive impairments [5,6].However, neurocognitive outcomes after chemotherapy treatment without cranial irradiation are inconsistent.Candidate mechanisms for cognitive impairment include direct neurotoxic effects causing atrophy of grey matter (GM) and/or demyelination of the white matter (WM), secondary immunologic responses leading to inflammatory reactions, microvascular damage, and genetic vulnerabilities [7][8][9].
In the past decades, the advent of neuroimaging techniques has provided insights into the anatomic Clinical Research Paper substrates of potentially toxic treatment effects.Studies on brain structural and functional alterations (including WM integrity [10][11][12], GM volume [13][14][15], restingstate magnetic resonance imaging (MRI) [16], task-state MRI [17][18][19], and cortical thickness [20,21]) related to ALL patients who take chemotherapy treatment has been accumulating in recent years.However, these studies have only focussed on regional brain changes, and little is known about global topological alterations specific to ALL with chemotherapy.
Currently, growing amounts of brain network studies have aimed to reconstruct whole-brain WM connectivity with diffusion tensor imaging (DTI) tractography based on graph theory [22,23].Graph theoretical analysis provides a powerful tool for quantifying the organization of network connectivity by defining a graph as a set of nodes (brain regions) and edges (structural connections) [23][24][25].These methods have revealed many important topological properties, such as small-world, which is characterized by a high global integration and a high local specialization between brain regions [26][27][28][29][30], and the network efficiency that characterizes the fault tolerance of the network [31].However, no studies have reported ALL with chemotherapy changes in WM brain networks or have investigated the topological differences between ALL patients and healthy controls.
In the present study, we used DTI tractography and graph theory to construct the WM structural networks in ALL patients with chemotherapy and healthy controls.Then, small-world properties and global parameters were compared between these two groups.We hypothesized that the network topographical structure might have been disrupted in ALL patients and hope to provide new insights into the mechanisms of chemotherapy-induced brain impairment.

Demographics
There were no significant differences in terms of age (two sample t-test, t = 1.433, p = 0.159) and gender (chi-square test, χ 2 = 1.067, p = 0.302) between the ALL patients and healthy controls (Table 1).

Small-worldness
At the threshold of three fibres, the structural networks of the ALL patients and healthy controls had higher normalized clustering coefficients (γ ALL = 4.462 ± 0.055, γ HC = 4.744 ± 0.088) but almost identical normalized shortest path lengths (λ ALL = 1.152 ± 0.011, λ HC =1.185 ± 0.024) relative to matched random networks, which suggested that the two groups exhibited typical small-world organizations in their structural networks.Statistical analysis indicated that ALL patients had a significant increase in the shortest path length Lp (P < 0.01) and significant decreases in the clustering coefficient C p (P < 0.01) and the normalized clustering coefficients γ (P < 0.01).No group differences were found for normalized shortest path length λ and smallworld σ (Figure 1 and Supplementary Table 1).

Network efficiency
We found significantly lower global efficiency Eg (P < 0.01) for ALL patients relative to healthy controls (Figure 1).No group difference was found for local efficiency Eloc (Figure 1).

Hubs
The regions were defined as network hubs if their nodal betweenness centrality was one standard deviation (SD) greater than the average of the network.The study identified 9 hub nodes of the WM structural networks in the healthy control group and 12 hub nodes in the ALL group.In both groups, 2 brain regions were identified as hubs in common, including the right precentral gyrus (PreCG.R) and the right middle frontal gyrus (MFG.R).In addition, some brain regions, including the left supplementary motor area (SMA.L), the right insula (INS.R), the left middle occipital gyrus (MOG.L), the bilateral putamen (PUT), the left precentral gyrus (PreCG.L), the right superior frontal gyrus_orbital (SFGorb.R), the right hippocampus (HIP.R), the right superior temporal gyrus (STG.R), and the left middle temporal gyrus (MTG.L) were identified as hubs in the ALL group but not in the healthy control group.Seven brain regions, including the right supplementary motor area (SMA.R), the left superior frontal gyrus_orbital (SFGorb.L), the left middle frontal gyrus (MFG.L), the left superior frontal gyrus_dorsolateral (SFGdor.L), the left rolandic operculum (ROL.L), the right precuneus (PCUN.R), and the left caudate nucleus (CAU.L), were identified as hubs in the healthy controls group but not in the ALL group (Figure 2).

DISCUSSION
In the present study, we explored whether there were aberrant topological properties in WM networks in ALL patients with chemotherapy treatment and healthy controls.Our main findings were as follows: a) smallworld organizations were present in both ALL patients with chemotherapy and controls, which is agreement with a previous study [42]; b) increased shortest path length Lp (P < 0.01), decreased in clustering coefficient C p (P < 0.01), and decreased normalized clustering coefficients γ (P < 0.01) were present in ALL patients; c) lower global efficiency Eg (P < 0.01) was present in ALL patients; d) hubs were located more in the parietal regions of healthy controls and in the temporal regions in ALL  patients; and e) alterations in nodal betweenness centrality were present in both groups.
Our study is the first to find that the ALL patients with chemotherapy treatment demonstrated reduced global connectivity in WM structure compared with healthy controls.This global hypo-connectivity in ALL patients with chemotherapy treatment is reflected by two abnormal indexes for small world measures (decreased γ and increased λ).As expected, the significantly decreased γ in ALL patients compared with healthy controls has been suggested to reduce global connectivity efficiency [23], which is in line with a previous study [42].Global efficiency is known to indicate the capacity to transfer information between different nodes of the brain network and is a comprehensive index for parallel information processing capabilities [39,43].Brain networks with small-world σ confer resilience against pathological attack, and impairments of these properties are caused by chemotherapy treatment in ALL patients.
We demonstrated a lower network efficiency in ALL patients with chemotherapy treatment compared with healthy controls, which suggested a less optimal organization for ALL patients.The significantly decreased C p and increased L p in ALL patients indicate the loss of the capacity to transmit specialized information rapidly among distant brain regions.It is suggested that the structural aberrations in some local brain areas might have contributed to the reduced clustering and longer path length in ALL patients compared with healthy controls.
There have been several studies on the morphological analysis of ALL patients with chemotherapy treatment [15,20,21,44], but the results have been inconsistent.Reduced thickness in the right precentral gyrus [21], decreased GM volume in several brain regions [15,20], and increased GM volume in the caudate [44] have also been reported.A possible explanation for these findings might be the mixing of various aetiologies, or the analyses were performed using different methods.By employing graph theoretical analysis for structural imaging, our study revealed the different structural connectivity networks between ALL patients with chemotherapy treatment and healthy controls.
The present study found that the hubs were mostly located in the parietal and temporal regions, which is compatible with former findings in which hubs were primarily located in parietal areas [41], reflecting the main functions associated with these areas [23].However, there was little overlap distribution of hubs between the two groups.Specifically, hubs in parietal regions accounted for a great proportion of all of the hubs in healthy controls, while more hubs were found in temporal regions in ALL patients with chemotherapy treatment.These findings suggest that parietal regions are more important in healthy populations, while temporal regions are more important in ALL patients.
A number of limitations need to be noted regarding the current study.First, the methodology of deterministic tractography to reconstruct structural connectivity cannot accurately map out all the fibres and has the limitation of tracking crossing fibres.Second, the sample size is limited, and the results need to be replicated in large samples.Third, this study included no neurocognitive testing; therefore, we cannot analyse the relationship between brain structure abnormalities and behaviour patterns.Finally, the present study is cross-sectional in design.Without a longitudinal study, the differential effects of cancer progression, chemotherapy, and/or treatmentrelated factors cannot be determined.A longitudinal study with a larger sample would be valuable to further define the network properties of ALL patients.
In conclusion, our study demonstrated the presence of altered topological properties in the structural networks of ALL patients with chemotherapy.In addition, the network metrics suggested that the efficiency of the entire brain network was decreased.The abnormality of the structural networks between ALL patients with chemotherapy and healthy controls is remarkable and could contribute to a better understanding of the effects of chemotherapy in ALL patients.

Participants
This study was approved by the Ethics committee of the Second Hospital of Anhui Medical University, and all participants obtained written informed consent.We identified 32 ALL patients who had completed all anti-cancer treatments with chemotherapy only in the Second Hospital of Anhui Medical University from June 2015 to June 2016.Four of the 32 survivors withdrew during data collection, leaving 28 participants (mean age: 40.71 ± 8.58 years; years since diagnosis: 7-38) included in the study.The ALL group had a history of chemotherapy treatment and was off-therapy for at least 3 years.Systemic chemotherapy included prednisolone, vincristine, cyclophosphamide, daunorubicin, doxorubicin, asparaginase, teniposide, cytarabine, 6-mercaptopurine, 6-thioguanine, and dexamethasone.Criteria for exclusion were as follows: (1) pre-existing neurologic disease affecting behaviour, (2) neurologic sequelae during treatment, (3) ALL relapse, (4) cranial irradiation or haematopoietic stem cell transplantation, and (5) MRI contraindication.For the control group, 20 healthy, agematched controls were selected.

Data preprocessing
Data preprocessing and network construction were performed using PANDA (www.nitrc.org/projects/panda),a pipeline toolbox for diffusion MRI analysis [32].Briefly, the image preprocessing steps included: format conversion of original data (DICOM), BET (skull removal), eddy current and head motion correction, fractional anisotropy (FA) calculation, and diffusion tensor tractography.

Network construction
Defining network nodes.Individual structural images were first co-registered to their first b0 images.Next, the transformed structural images were nonlinearly normalized to the MNI space.Finally, the derived deformation parameters were inverted and employed to warp the automated anatomical labelling (AAL) atlas [33] from the MNI space to the diffusion image native space.After this procedure, we obtained 90 cortical and subcortical regions (45 for each hemisphere), each representing a network node [23].
Defining network edges.To define the edges of the structural network, we calculated the number of fibres (with end-points in both nodes during the fibre tracking) between two regions, generating a 90 × 90 matrix for each subject (Figure 3).To reduce false-positive connections due to the limited resolution of DTI when a minimum of three fibres is used as a threshold [34][35][36], the fibre number (FN) between each couple of brain nodes was defined as the weights of the network edges.Finally, the FN-weighted structural networks were constructed for each participant [37,38].

Network analysis
The WM network topological properties were analysed using graph theory in GRETNA (www.nitrc.org/projects/gretna).For the structural-weighted brain network, we calculated the network metrics at the global level.We mainly focussed on the clustering coefficient Cp, shortest path length Lp, small-worldness (normalized clustering coefficient γ and normalized shortest path length λ) [26], local efficiency Eloc, and global efficiency Eg.The Cp of the entire network was the average of that of all nodes in the network, which refers to how the neighbouring nodes and the node itself are connected and indicates the local efficiency for the transformation of the information [23].The Lp of the entire network is the average of that of all nodes in the network, which represents the average shortest travel distance between another node and the node itself and indicates the most efficient information transfer between the two nodes.To obtain a normalized clustering coefficient γ and a normalized shortest path length λ, we generated 100 matched random networks that had the same numbers of nodes and edges and degree distributions but preserved the weighted distribution of the real network.We calculated the average Cp random and Lp random over the random network.Then, we computed the γ (γ = Cp/Cp random ) and λ (λ = Lp/Lp random ), which indicate the normalized clustering coefficient and normalized shortest path length, respectively.In a small-world network, the γ should be much larger than 1, and the λ should be close to 1.
The global efficiency Eg is the average of the inverse of the shortest path length of all node pairs in the network and usually represents the capability of the network for performing parallel information processing.The local efficiency Eloc is the average of the global efficiency of the community neighbouring all nodes in the network and reflects the fault tolerance level of the network [39,40].

Identification of hubs
Hubs of structural networks are essential nodes that are identified in various ways [41].In the present study, we applied betweenness to identify the hubs of the network.A node was considered as a hub if its regional betweenness centrality was one SD higher than the mean network betweenness.

Statistical analysis
Statistical analysis was performed with SPSS 16.00 software.The group difference in age was tested using the two-sample t-test and the gender ratio was examined with a Pearson chi-square test.For group differences in global network measures (Cp, Lp, γ, λ, σ, Eloc, Eg), was used the two-sample t-test.

Figure 1 :
Figure 1: Differences in global metrics of the structural networks between ALL and HC groups.Error bar represent standard errors.Cp, clustering coefficient; Lp, shortest path length; Gamma, normalized clustering coefficient; Lambda, normalized shortest path length; sigma, small-worldness; E g , global efficiency; E loc , local efficiency.ALL, acute lymphoblastic leukemia with chemotherapy treatment, HC, healthy controls.

Figure 3 :
Figure 3: Structural connectivity (fiber number) matrices for the construction of structural weighted networks for ALL (A) and HC (B).The axes represent the 90 AAL brain regions used in the analysis.Color of each entry represents the level of connectivity.ALL, acute lymphoblastic leukemia with chemotherapy treatment, HC, healthy controls.