Abstract

   DNA methylation levels are associated with neurodevelopment.
   Attention-deficit/hyperactivity disorder (ADHD), characterized by
   attention deficits, is a common neurodevelopmental disorder. We used
   methylation microarray and pyrosequencing to detect peripheral blood
   DNA methylation markers of ADHD. DNA methylation profiling data from
   the microarray assays identified potential differentially methylated
   CpG sites between 12 ADHD patients and 9 controls. Five candidate CpG
   sites (cg00446123, cg20513976, cg07922513, cg17096979, and cg02506324)
   in four genes (LIME1, KCNAB2, CAPN9, and SPTBN2) were further examined
   with pyrosequencing. The attention of patients were tested using the
   Conners’ Continuous Performance Test (CPT). In total, 126 ADHD patients
   with a mean age of 9.2 years (78.6% males) and 72 healthy control
   subjects with a mean age of 9.3 years (62.5% males) were recruited.
   When all participants were categorized by their CPT performance, the
   DNA methylation levels in LIME1 (cg00446123 and cg20513976) were found
   to be significantly higher and those in SPTBN2 (cg02506324) were
   significantly lower in children with worse CPT performance. Therefore,
   DNA methylation of two CpG sites in LIME1 and one CpG site in SPTBN2 is
   associated with attention deficits in children. DNA methylation
   biomarkers may assist in identifying attention deficits of children in
   clinical settings.

   Keywords: ADHD, epigenetic, neuropsychological test, DNA methylation,
   biomarker

1. Introduction

   Attention-deficit/hyperactivity disorder (ADHD) is a common
   neuropsychiatric disorder that is characterized by inattention, and/or
   excessive activity and impulsivity, [[36]1]. This disorder affects
   approximately 5% of school-age children globally [[37]2]. ADHD is
   generally considered to be a highly hereditary disease [[38]3] and is a
   candidate disease for understanding the pathophysiology that underlies
   attention deficits. Epigenetic regulation refers to gene regulation
   that changes the phenotype without altering nucleotide sequence
   [[39]4,[40]5]. DNA methylation is a chemical modification of DNA that
   changes genetic expression without altering the DNA sequence and is a
   mechanism of gene regulation [[41]6]. DNA methylation typically occurs
   in CpG dinucleotides, and the regions in which CpG aggregates are
   called CpG islands, such as the promoter region in the first exon.
   Previous studies have reported that prenatal stress could alter the
   brain function in offspring [[42]7], and maternal diet could result in
   DNA methylation changes in offspring [[43]8]. Moreover, DNA methylation
   changes may be associated with prenatal exposure to maternal anxiety or
   cigarette smoking [[44]9,[45]10].

   DNA methylation that is related to inflammatory responses and the
   neuronal conduction of monoamines has been associated with ADHD
   [[46]11,[47]12]. A differential DNA methylation marker has been
   identified in VIPR2 and MYT1L genes for ADHD [[48]13,[49]14]. The
   degree of methylation of the serotonin transporter gene (SLC6A3 and
   SLC6A4) promoter is negatively correlated with the thickness of the
   right temporal lobe and the occipital cortex in ADHD patients and
   positively correlated with the impulsive symptoms of such patients
   [[50]15,[51]16,[52]17]. Numerous studies have also reported that the
   degree of methylation of DRD4 is significantly correlated with ADHD
   severity [[53]18,[54]19,[55]20,[56]21]. Rijlaarsdam et al. [[57]22]
   pointed out that in cases of early-onset and persistent behavioral
   disorders, a prenatal unhealthy diet is related to more severe ADHD
   symptoms via higher IGF2 methylation. In addition, many genes and
   pathways were reported to be implicated in attention or
   intelligence-related disorders, including NISCH, ST3GAL3 [[58]23],
   COMT, IFNG, MIR200B, SYNGAP1 [[59]24,[60]25], BDNF and NGFR [[61]26]. A
   prospective meta-analysis which contained 2374 children DNA methylation
   (cg25520701) lies within the gene body of CREB5 was associated with
   neurite outgrowth and an ADHD diagnosis [[62]27].

   Methylome-wide analysis is a hypothesis-free methodfor determining
   which genes are involved in the disease’s pathogenesis, which has the
   potential to find novel biological associations [[63]28]. Several
   studies have used the Illumina Human Methylation 450K (Illumina 450k)
   array as a tool for methylome-wide analysis to detect ADHD methylomic
   biomarkers. Walton et al. [[64]29] reported that DNA methylation across
   multiple genomic locations at birth differentiated ADHD trajectories.
   For example, SKI is involved in neural tube development; ST3GAL3 is
   related to intellectual disability; PEX2 is linked to peroxisomal
   processes, and ZNF544 is implicated in ADHD. Heinrich et al. [[65]26]
   analyzed methylation patterns in buccal cell DNA using the Illumina
   450K array. CpG sites in the genes of the dopaminergic (COMT, ANKK1)
   and the neurotrophic (BDNF, NGFR) system are associated with Nogo-P3
   and the severity of ADHD symptoms. Effects of methylation are related
   to both the functional aspects of cognitive control, motivation, and
   ADHD-related behaviors. Cecil et al. [[66]30] examined genome-wide DNA
   methylation patterns using buccal swabs and identified themethylation
   levels of the DRD4 gene as a potential biomarker of aggressive
   behavior.

   The currently available genome-wide Illumina Infinium Human Methylation
   850 BeadChip array (M850K) allows researchers to identify more than
   850,000 methylation sites in a genome at the resolution of a single
   nucleotide and increases their ability to detect novel methylomic
   markers. No investigation has yet used the Illumina M850K to identify
   DNA methylation biomarkers of attention deficits. We performed the
   power calculation (α = 0.05, β = 0.8, allocation ratio N1/N2 = 1.5) and
   sample size would be 80 for patients and 54 for controls for detecting
   a moderate effect (0.5). We propose that DNA methylationsites
   identified using the M850K system can be a potential biomarker for
   detecting attention deficits among children. This work aims to identify
   potential DNA methylation biomarkers and to determine whether
   differentially expressed DNA methylation is associated with the
   characteristics of ADHD patients.

2. Materials and Methods

2.1. Study Participants

   The research protocol (104-9250A3) was approved by the Institutional
   Review Board (IRB) at Chang Gung Memorial Hospital in Taiwan, and
   written informed consent was obtained from the parents or guardians of
   all participating children. The subjects of this study were eligible
   patients with ADHD treated in the outpatient Department of Child
   Psychiatry at Chang Gung Children’s Hospital in Taiwan along with
   healthy control children. The methods used in our previous research
   [[67]31] were adopted. The criteria for selecting the ADHD patients
   included the following: (a) a clinical diagnosis of ADHD by a senior
   child psychiatrist based on the criteria provided in the Diagnostic and
   Statistical Manual of Mental Disorders, Fourth Edition, Text Revision
   (DSM-IV-TR), following structured interviews based on the Chinese
   version of the Schedule for Affective Disorders and Schizophrenia for
   School-Age Children, epidemiologic version (K-SADS-E) [[68]31]; (b)
   kids aged between 6 and 16 years; (c) a Han Chinese ethnic background;
   (d) kids who have never taken any medications to treat ADHD. Patients
   suffering from major physical illnesses (such as genetic, metabolic, or
   infectious conditions) or having a history of comorbid major
   neuropsychiatric diseases (such as intellectual disabilities, autism
   spectrum disorder, bipolar disorders, major depressive disorders,
   psychotic disorders, substance use disorders, epilepsy, or severe head
   trauma) were excluded.

   The control subjects consisted of healthy children undergoing health
   checkups and blood draws for laboratory examinations. The remaining
   blood samples of these subjects were applied for this study. The
   healthy control subjects were children who do not suffer from ADHD, are
   ethnically Han Chinese, between the ages of 6 and 16 years old, and
   from within the same catchment area. They have not suffered from any
   known major physical illness or any of the major neuropsychiatric
   diseases mentioned above.

2.2. Collecting Clinical Blood Samples and M850K Examination

   Approximately 5 mL of blood was collected from each subject. Whole
   blood samples were placed in a centrifuge (3000 rpm for 10 min) and a
   red blood cell (RBC) lysis buffer (RBCBioscience) was added to remove
   the RBCs. Then, we collected DNA samples from the enriched total white
   blood cells (WBCs). The collected DNA samples were then subjected to
   bisulfite conversion with EZ DNA Methylation-Lightning ^TM Kit (Zymo
   Research, Irvine, USA). Next, the bisulfite converted DNA samples were
   subjected to Infinium Human Methylation 850 BeadChip array (M850K,
   Illumina, CA, USA).

   The generated raw M850K data were first processed with GenomeStudio
   (Illumina, CA, USA) with the default parameters as suggested by a
   previous study [[69]32]. Then, the GenomeStudio output data were
   subsequently analyzed with Partek Genomics Suite (Qiagen, CA, USA).
   Partek Genomics Suite first inputted the microarray data, followed by
   normalizing the data based on the Functional normalization method
   proposed by a previous study [[70]33]. We had the samples divided into
   two sets, control, and disease. By comparing the two sets, we conducted
   one-way ANOVA and requested the criteria of a significant CpG
   dinucleotide as follows: FDR < 0.05 and beta value variation > 1.2. The
   FDR values were calculated based on the method proposed by Benjamini et
   al. ([71]https://doi.org/10.1111/j.2517-6161.1995.tb02031.x). We also
   deal with the batch effect by specifying the “Remove Batch Effect”
   function of Partek. In addition, we also removed the CpG dinucleotides
   with missing beta values. We further also mapped the significant CpG
   dinucleotides back to the human genome (hg19) and determined whether
   they were located with the promoter regions (−5000 to +3000 base pair
   from transcription start site) based on the annotation of RefSeq 82.

2.3. Pyrosequencing and Sequence Analysis

   DNA from WBC obtained from blood samples was used for pyrosequencing
   and sequence analysis. Pyrosequencing was conducted using the PSQ Gold
   SQA reagent kit (Qiagen) on a Pyromark Q96 ID platform (Qiagen, Hilden,
   Germany), according to the manufacturer’s instructions. Simply put, 50
   μL of biotinylated PCR products were mixed with 3 μL of streptavidin
   sepharose beads (Amersham Biosciences AB, Umeå Västerbotten, Sweden)
   and 47 μL of binding buffer, and the mixture was then shaken at 2000
   rpm for 10 min. A vacuum preparation tool was used to capture the
   immobilized complex. The single-strand purification was carried out to
   separate and discard the unbiotinylated strand. The beads that were
   bound to single biotinylated strands were released to a 96-well
   microtiter plate to which 49 μL of annealing buffer and 1 μL of
   complementary sequencing primer (pBR-V1.AS) had been added. The
   processed mixture was loaded onto the PyroMark ID system for
   pyrosequencing with 10 cycles of ATCG dispensing.

2.4. Neuropsychological Assessments

   Conners’ Continuous Performance Test (Conners CPT) is a 14-min
   computerized test that primarily assesses attention and impulse
   control. [[72]34] Briefly, participants are required to respond to
   stimuli on a computer screen by pressing a space bar whenever they see
   a letter that is not “X.”Several parameters are assessed. The
   Confidence Index (percentile) incorporates all CPT data that are
   obtained in the test in a probability stated as a number out of 100
   that a significant attention problem exists. A lower T-score on the CPT
   represents better performance [[73]35]. In this study, we categorized
   patients based on the Confidence Index of 50 into the better
   performance group (CPT ≤ 50) and the worse performance group (CPT >
   50).

2.5. Clinical Measurements

   A senior psychiatrist used the K-SADS-E diagnostic tool [[74]31] to
   interview all participants in both the ADHD patient group and the
   control group. An experienced child psychologist used the Wechsler
   Intelligence Scale for Children–Fourth Edition (WISC-IV) [[75]36] to
   evaluate individual patients in a room that was designed to reduce
   variability in testing conditions. The parents and a teacher of the
   patients were asked to complete the Swanson, Nolan, and Pelham Version
   IV Scale (SNAP-IV) parent form and SNAP-IV teacher form [[76]37],
   respectively.

2.6. Statistical Analysis

   Data were analyzed using the statistical software package SPSS, version
   16.0 (SPSS Inc., Chicago, IL, USA). Variables were presented in terms
   of either mean (standard deviation) or frequency. Two-tailed p-values
   of < 0.05 were considered to indicate statistical significance. The
   Chi-square test or Fisher’s exact test was used to compare the gender
   distributions of the ADHD patients and the controls and those of the
   groups with CPT > 50 and CPT ≤ 50. Multivariate Analysis of Covariance
   (MANCOVA) was used to determine the potential difference in DNA
   methylation dinucleotides between the ADHD group and control, while the
   effect of age, sex, and intelligence quotient (IQ) were put under
   control.

3. Results

3.1. DNA Methylation Profile

   We had 9 control and 12 disease samples (age-and gender-matched)
   analyzed with M850K assays. After the GenomeStudio data was loaded,
   Partek initially generated the PCA plot. As shown in [77]Figure 1a, the
   disease samples aggregated at a small space; while, the control samples
   scattered at a larger space. Such a result implied that the disease
   samples had more similar DNA methylation patterns than control samples
   did. [78]Figure 1a also demonstrated that one control sample was closer
   to the disease samples than to the remaining control ones, which
   implied that it was more similar to the disease samples in terms of a
   DNA methylation pattern.

Figure 1.

   [79]Figure 1
   [80]Open in a new tab

   The overall DNA methylation patterns were derived from the M850K
   analysis. Blue and red dots (bars) denoted the control (healthy
   control) and disease (ADHD) subjects, respectively. (a) The PCA plot
   reflected the overall DNA methylation patterns of samples. The disease
   samples were aggregative more than the control samples. (b) The heat
   map reflected the variations of the significant CpG dinucleotides.
   Almost 95% of significant CpG dinucleotides were hypo-methylated in the
   disease set.

   Under our criteria (FDR < 0.05 and beta variation > 1.2), 14,626 CpG
   dinucleotides were determined as significant based on ANOVA. Their
   variations (hyper-methylation or hypo-methylation) were demonstrated
   with the heat map ([81]Figure 1b). [82]Figure 1b demonstrated that the
   most significant CpG dinucleotides were hypo-methylated in disease
   samples and such result was pretty similar to previous studies. In
   inflammation-related diseases, such as Kawasaki disease [[83]38] or
   autoimmune encephalitis [[84]32], significant CpG dinucleotides tend to
   be hypo-methylated in disease samples. [85]Figure 1b also illustrated
   that one control sample was clustered into the disease set which was
   consistent with the result of the PCA plot. The overall results of CpG
   methylation were shown with a Manhattan plot in [86]Supplementary
   Figure S1.

   By referring to the RefSeq 82 data, these significant 14,626 CpG
   dinucleotides were located at the putative promoter regions of 9749
   genes. We also conducted pathway enrichment analysis on the 9749 genes
   ([87]Supplementary Table S1). However, no obvious neuron-related
   pathway was enriched. Therefore, we investigated whether the genes
   hosted by significant CpG dinucleotides were associated with
   neurobiological processes, neurodevelopment, neurotransmission, or
   neuroprotection. As a result, a total of four genes (LIME1, KCNAB2,
   CAPN9, and SPTBN2) and five CpG sites (cg00446123, cg20513976,
   cg07922513, cg17096979, and cg02506324) were selected for further
   pyrosequencing analysis.

3.2. Pyro-Sequencing Analysis Result

   [88]Table 1 presents the characteristics of the ADHD patients and the
   healthy controls. A total of 126 ADHD patients (mean age 9.2 years,
   78.6% males) and 72 healthy control subjects (mean age 9.3 years, 62.5%
   males) were recruited. ADHD patients were more likely than the control
   subjects to be male (p = 0.015) and to have a lower intelligence
   quotient (p < 0.001), higher inattention scores (p < 0.001), and higher
   hyperactivity/impulsivity scores (p < 0.001) as rated by parents or
   teachers; they also had worse values of the CPT performance, including
   Confidence Index (p < 0.001), Omission (p = 0.002), Commission (p =
   0.039) and Detectability (p = 0.036) in the CPT. With respect to the
   four target genes (LIME1, KCNAB2, CAPN9, and SPTBN2) and the five CpG
   sites (cg00446123, cg20513976, cg07922513, cg17096979, and cg02506324),
   no significant difference in the levels of the CpG sites were
   identified between the ADHD patients and the controls ([89]Table 2).

Table 1.

   Characteristics of Patients with ADHD and Healthy Controls.
   Characteristics ADHD
   (N = 126) Controls
   (N = 72) Statistic p-Value
   Sex 5.967 0.015 *
   Male 99 (78.6) 45 (62.5)
   Female 27 (21.4) 27 (37.5)
   Age(years) 9.2 ± 2.1 9.3 ± 2.2 0.342 0.733
   FSIQ of the WISC-IV 97.1 ± 10.5 108.3 ± 13.7 5.960 <0.001 *
   Clinical measures
   SNAP-IV parent form (I) 16.4 ± 5.5 5.8 ± 6.0 12.921 <0.001 *
   SNAP-IV parent form (H) 14.3 ± 6.5 4.5 ± 5.5 10.916 <0.001 *
   SNAP-IV teacher form (I) 14.8 ± 5.9 4.5 ± 4.9 13.188 <0.001 *
   SNAP-IV teacher form (H) 11.7 ± 6.9 2.9 ± 3.5 11.844 <0.001 *
   CPT
   Confidence Index 62.5 ± 22.6 46.7 ± 21.0 4.966 <0.001 *
   Omission 58.8 ± 18.1 51.6 ± 13.8 3.142 0.002 *
   Commission 49.2 ± 9.8 46.2 ± 10.5 2.079 0.039 *
   Hit Reaction Time 55.4 ± 12.9 55.5 ± 11.4 0.012 0.990
   Detectability 51.2 ± 9.1 48.1 ± 10.9 2.109 0.036 *
   [90]Open in a new tab

   Data are expressed as N (%) or Mean ± SD; FSIQ, Full-Scale Intelligence
   Quotient; WISC-IV, Wechsler Intelligence Scale for Children–Fourth
   Edition; I, inattention scores; H, hyperactivity/impulsivity scores; *
   p < 0.05.

Table 2.

   DNA Methylation Levels of Patients with ADHD and Healthy Controls.
   Gene CpG ADHD (N = 126)
   Mean (SD) Controls (N = 72)
   Mean (SD) F p-Value
   LIME1 cg00446123+5 57.9 ± 9.4 58.9 ± 7.6 0.110 0.740
   LIME1 cg00446123+9 50.6 ± 8.8 50.2 ± 7.8 0.054 0.817
   LIME1 cg20513976 51.9 ± 8.0 51.6 ± 7.0 0.000 0.990
   LIME1 cg20513976+5 58.6 ± 9.4 59.8 ± 8.2 0.832 0.363
   LIME1 cg20513976+9 50.6 ± 9.0 50.0 ± 7.6 0.025 0.875
   KCNAB2 cg07922513 49.5 ± 6.7 48.9 ± 5.7 0.295 0.588
   CAPN9 cg17096979 78.6 ± 15.4 81.6 ± 15.7 0.963 0.328
   SPTBN2 cg02506324 35.5 ± 5.5 35.2 ± 6.2 0.134 0.714
   [91]Open in a new tab

   Owing to the design of the M850K microarray assay, each probe detects
   only one CpG dinucleotide. In addition, the neighboring probes keep a
   moderate distance to avoid mutual interference, which leaves many CpG
   dinucleotides locating between probes undetected. In this study, we
   used PyroMark to validate the methylation profiles of specific CpG
   dinucleotides by sequencing ~50 nt fragments. Therefore, the
   neighboring originally un-detected CpG dinucleotides could also be
   validated. For example, cg20513976+5 and cg20513976+9 were 5 and 9 base
   downstream apart from cg20513976, respectively.They were not detected
   with M850K assay but validated by PyroMark assay simultaneously with
   cg20513976. Data are expressed as F value and p-value, using MANCOVA,
   and controlling for age, sex, and Full Scale Intelligence Quotient of
   WISC-IV, Wechsler Intelligence Scale for Children–Fourth Edition.

   In our exploratory analyses, we further examined the levels of the CpG
   sites by comparing the CPT difference. [92]Supplementary Table S2
   summarized the characteristics of children with worse CPT performance
   (n = 115) and those with better CPT performance (n = 83). The group
   with the worse CPT group more likely to be male (p = 0.002), and to
   have higher inattention scores (p = 0.027) and higher
   hyperactivity/impulsivity scores (p = 0.018) as rated by parents,
   higher inattention scores (p = 0.003), and higher
   hyperactivity/impulsivity scores (p = 0.037) as rated by teachers, and
   worse values of the performance Confidence Index (p < 0.001), Omission
   (p < 0.001), Hit Reaction Time (p < 0.001) in the CPT.

   After controlling for age, sex, or IQ ([93]Supplementary Table S3), DNA
   methylation levels in LIME1 (cg00446123 and cg20513976) were
   significantly higher, and those in SPTBN2 (cg02506324) were
   significantly lower, in children with worse performance in the CPT. DNA
   methylation levels in KCNAB2 (cg07922513) and CAPN9 (cg17096979) did
   not differ significantly between children with worse and those with
   better CPT performance. We also tried a sex-stratified analysis and
   found that DNA methylation levels in LIME1 (cg00446123) were
   significantly higher (p-value = 0.039) in boys with worse performance
   in the CPT. However, no significant difference in DNA methylation
   levels between different CPT performance and diagnosis group among
   girls.

4. Discussion

   This work applied the M850K system to identify candidate DNA
   methylation markers for ADHD. A previous large epigenome-wide
   association study of childhood ADHD revealed peripheral DNA methylation
   (cg17478313 annotated to SLC7A8 and cg21609804 annotated to MARK2)
   associated with disease and polygenic risk burden [[94]39]. The study
   herein identified two novel DNA methylations, involving LIME1
   (cg00446123 and cg20513976) and SPTBN2 (cg02506324), which are
   identified as effective in identifying attention deficits in children.

   The Lck interacting transmembrane adaptor 1 (LIME1) gene encodes a
   transmembrane adaptor protein that links the T and B-cell receptor to
   downstream signaling pathways through its association with the Src
   family kinases Lck and Lyn, respectively. Splicing of this gene yields
   multiple transcript variants [[95]40]. The transcription start site of
   the LIME1 gene, which has a key role in inflammatory pathways such as
   MAPK signaling, has been implicated in the pathophysiology of cerebral
   palsy [[96]41]. The immune system may be affected by common genetic
   background and molecular mechanisms [[97]42]. Prenatal exposure to
   inflammation is associated with changes in brain development in young
   offspring, including reductions in cortical gray matter volume and the
   volume of certain cortical areas [[98]43]. The pathophysiology of ADHD
   is involved in morphological and functional brain abnormality, and
   modifications to neurotransmitter systems, including the dopaminergic,
   serotonergic, and glutamatergic systems [[99]44]. Therefore, DNA
   methylation of the LIME1 (cg00446123 and cg20513976) gene may influence
   that gene’s function and further affect the corresponding inflammatory
   pathway, which may be linked to the phenotype of attention deficits.

   Spectrins are principal components of a cell’s membrane-cytoskeleton
   and comprise two alpha and two beta spectrin subunits. Spectrin beta
   non-erythrocytic 2 (SPTBN2) regulates the glutamate signaling pathway
   and is associated with neonatal cerebellar cortical degeneration
   [[100]45]. Heterozygous missense variants in the SPTBN2 gene, encoding
   the non-erythrocytic beta spectrin 2 subunit (beta-III spectrin), have
   been identified in autosomal dominant spinocerebellar ataxia type 5
   (SCA5), which is a rare adult-onset neurodegenerative disorder that is
   characterized by progressive cerebellar ataxia
   [[101]46,[102]47,[103]48,[104]49]. The result herein reveals a
   potential association between SPTBN2 and neurodevelopment. DNA
   methylation at the CpG sites of the SPTBN2 gene may influence the
   process of neurodevelopment and be further linked to attention.

   The interesting issue is that DNA methylation levels were not directly
   associated with ADHD diagnosis, but highly correlated with attention
   deficits as measured by the CPT. A diagnosis of ADHD depends on
   gathering information about behavioral symptoms from patients and their
   parents, teachers, or clinical observers. However, interviews and data
   collection are time-consumingand behavioral measures are vulnerable to
   rater and reporting biases [[105]50]. The CPT is a computer-based
   neuropsychological test that is commonly used to measure an
   individual’s attention and impulsivity during a sustained task; it can
   be used in a clinical inquiry as part of a diagnostic procedure
   [[106]51]. Numerous investigations have suggested that combining CPTs
   with an objective measure of activity may be particularly effective for
   clinical purposes [[107]52,[108]53]. We previously found that the
   association between levels of ADHD biomarkers and CPT performance was
   stronger than that with the severity of behavioral symptoms
   [[109]54,[110]55]. Consistent with the findings herein, the CPT may
   serve as an instrument for detecting the biological footprint and
   underlying pathophysiology of ADHD [[111]56].

   Studies have shown that the degrees of DNA methylation in peripheral
   blood and brain tissue are highly correlated [[112]57]. Previous
   investigations have found DNA methylation markers in various genes,
   including the serotonin transporter gene (SLC6A3) [[113]15], SLC6A4
   [[114]16], DRD4, 5-HTT
   [[115]17,[116]18,[117]19,[118]20,[119]21,[120]30], VIPR2, MYT1L
   [[121]13], 5-HT3A [[122]17], IFNG [[123]24], COMT, ANKK1, BDNF, NGFR
   [[124]9], NISCH and ST3GAL3 [[125]23] genes. However, the
   aforementioned DNA methylation markers were not identified in this
   study. This discrepancy has numerous explanations. The first is that in
   previous studies, an illumine M450K system was used as an explorative
   tool as part of a target gene strategy, based on neurobiological
   theory. In contrast, this study is the first in which M850K is used to
   detect DNA methylation markers, and the fold change between patients
   and controls in our preliminary study was not large enough for
   detecting these markers. Second, some previous studies used saliva or
   other samples, whereas the DNA methylation information herein was
   obtained from WBC. DNA methylation may be tissue-specific, leading to
   variations among studies.

   This investigation has several limitations. First, the
   HumanMethylation850K array captures a part of all CpGs in the genome.
   These polymorphisms reduce the quality of microarray-based measurements
   of DNA methylation and may limit the detection for DNA
   methylation–mediated effects of SNP on the phenotype. Second, the
   sample was small, and attention deficits co-occurred with the male
   gender. ADHD is generally heterogeneous and can be categorized into
   different subtypes, and psychiatric comorbidities can be observed in
   most ADHD patients. Owing to the smallness of the sample, whether
   gender or the heterogeneity of ADHD affects the results of DNA
   methylation was not addressed. Moreover, a MANCOVA analysis was used
   instead of a conditional logistic regression or logistic regression
   which may underestimate the standard errors [[126]58]. Third, the DNA
   methylation information was obtained using total WBC. Although the
   degrees of DNA methylation in peripheral blood and brain tissue could
   be highly correlated [[127]57], blood is unlikely to provide a complete
   picture of ADHD-related epigenetic processes, particularly in the
   brain. That is why the pathway or GO analyses based on the genes hosted
   by the significant CpG dinucleotides did not identify the
   neuron-related functions [[128]27], including this study
   ([129]Supplementary Table S3). Actually, the leukocyte population is
   composed of many different types of cells, including neutrophil,
   eosinophil, basophil, lymphocyte, and monocyte. In addition, lymphocyte
   can be further classified into b cell, CD4+ T helper cell, CD8+
   cytotoxic T cell, regulatory T cell, natural killer cells, and so on.
   Recently, single-cell sequencing technology was invented to provide
   higher resolution into cell-type-specific patterns [[130]59]. With the
   application of single-cell technology combined with bisulfite
   sequencing, we may enhance our understanding from total WBC level into
   leukocyte cell-specific level. By doing so, the methylation pattern
   from blood could better reflect disease. Fourth, the associations that
   were detected in this cross-sectional study may provide insights into
   the underlying biological process but not into causality. Future
   longitudinal studies of humans will help to overcome this issue.
   Finally, the participants in this study were all Han Chinese and
   recruited from a single hospital in Taiwan. Further research must be
   performed to determine whether our findings concerning DNA methylation
   markers can be generalized to various ethnicities or countries.

5. Conclusions

   This work newly finds that DNA methylation at two CpG sites in LIME1
   (cg00446123 and cg20513976) and one CpG site in SPTBN2 (cg02506324) is
   associated with attention deficits in children. DNA methylation
   biomarkers, while not associated with the diagnosis of ADHD, may help
   to identify attention deficits in clinical settings. Future studies are
   required to clarify the modulation of gene functions by DNA methylation
   and the biological mechanisms that underlie the attention modality.

Acknowledgments