Available via license: CC BY 4.0
Content may be subject to copyright.
Title: Human genomic regions of systemic interindividual epigenetic variation
are implicated in neurodevelopmental and metabolic disorders
Authors: Wen-Jou Chang1†, Uditha Maduranga1, Chathura J. Gunasekara1, Alan Yang1,
Matthew Hirschtritt2,3,4, Katrina Rodriguez5, Cristian Coarfa6,7, Goo Jun8, Craig L. Hanis8, James
M. Flanagan9, Ivan P. Gorlov10,11, Christopher I. Amos7,10,11, Joseph L. Wiemels12,13, Catherine
A. Schaefer4, Ezra S. Susser14, Robert A. Waterland1,15*
Affiliations:
1USDA/ARS Children’s Nutrition Research Center, Department of Pediatrics, Baylor
College of Medicine; Houston, TX, USA.
2Department of Psychiatry, Kaiser Permanente Oakland Medical Center; Oakland, CA, USA.
3Department of Psychiatry and Behavioral Sciences, University of California, San Francisco;
San Francisco, CA, USA.
4Division of Research, Kaiser Permanente Northern California; Pleasanton, CA, USA.
5Division of Behavioral Health Services and Policy Research, New York State Psychiatric
Institute; New York, NY, USA.
6Department of Molecular and Cellular Biology, Baylor College of Medicine; Houston, TX,
USA.
7Dan L Duncan Comprehensive Cancer Center, Baylor College of Medicine; Houston, TX,
USA.
8Human Genetics Center, Department of Epidemiology, University of Texas Health Science
Center; Houston, TX, USA.
9Division of Cancer, Department of Surgery and Cancer, Faculty of Medicine, Imperial
College London; London, UK.
10Institute for Clinical and Translational Research, Baylor College of Medicine; Houston,
TX, USA.
11Department of Medicine, Section of Epidemiology and Population Sciences, Baylor
College of Medicine; Houston, TX, USA.
12Center for Genetic Epidemiology, Department of Population and Public Health Sciences,
University of Southern California Keck School of Medicine; Los Angeles, CA, USA.
13USC Norris Comprehensive Cancer Center, University of Southern California; Los
Angeles, CA, USA.
14Mailman School of Public Health, Department of Epidemiology and Psychiatry, Columbia
University; New York, NY, USA.
15Department of Molecular & Human Genetics, Baylor College of Medicine; Houston, TX,
USA.
*Corresponding Author: Robert A. Waterland. Email: waterland@bcm.edu
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
NOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.
Abstract: Epigenome-wide association studies (EWAS) profile DNA methylation across the
human genome to identify associations with diseases and exposures. Most employ Illumina
methylation arrays; this platform, however, under-samples interindividual epigenetic variation.
The systemic and stable nature of epigenetic variation at correlated regions of systemic
interindividual variation (CoRSIVs) should be advantageous to EWAS. Here, we analyze 2,203
published EWAS to determine whether Illumina probes within CoRSIVs are over-represented in
the literature. Enrichment of CoRSIV-overlapping probes was observed for most classes of
disease, particularly for neurodevelopmental disorders and type 2 diabetes, indicating an
opportunity to improve the power of EWAS by over 200- and over 100-fold, respectively. EWAS
targeting all known CoRSIVs should accelerate discovery of associations between individual
epigenetic variation and risk of disease.
Main Text:
Genomic studies including genome-wide association studies (GWAS) and whole-genome and
exome sequencing have expanded our understanding of complex diseases by enabling
identification and characterization of genetic variants associated with specific diseases and traits
(1-6). Despite this success, for most common diseases the genetic variants identified by genomic
studies explain only a fraction of the risk (7), suggesting a role for individual epigenetic variation
in disease susceptibility (8, 9). For over a decade, epigenome-wide association studies (EWAS)
have focused on profiling CpG methylation at the genome scale to uncover links between
epigenetic variation and diseases or phenotypes (10, 11). Although thousands of studies have been
conducted, there are few robust and replicable associations. Recent reviews summarizing EWAS
conducted using the Illumina methylation platform (12-14) reveal poor replicability across studies
encompassing a range of diseases and outcomes, leading some to question if epigenetic variation
is indeed a major factor in disease risk (15).
An important consideration, however, is that the Illumina methylation platform, the workhorse of
EWAS, was not designed for detection of population level associations. The success of genomic
studies, in particular GWAS, was built upon the HapMap (16) and 1000 Genomes (17) projects,
which identified sites of common sequence variation in the human genome so they could be
assayed at the population level. The EWAS era, however, was launched without an analogous
project to map out human genomic regions of interindividual epigenetic variation. Instead,
investigators adopted the Illumina methylation array platform, principally the HM450 and then the
EPIC 850 array, which were designed to detect tissue- and gene-level DNA methylation profiles.
As these arrays were designed without knowledge of interindividual epigenetic variation (18, 19),
~90% of the probes (depending on the specific criteria) target CpG sites that show negligible
interindividual variation in normal somatic tissues (20-23), compromising their ability to detect
associations. Indeed, the low interindividual variation of most probes on the Illumina methylation
platform is associated with low technical reliability (20, 21) and poor stability in test-retest
experiments over time (22, 23).
Whereas the need for interindividual variation is common to GWAS and EWAS, factors unique to
EWAS also warrant consideration (9). Unlike DNA sequence, DNA methylation is largely cell
type-specific (24). Nonetheless, most EWAS conducted so far have profiled methylation in blood
DNA, even when attempting to draw associations with diseases whose pathophysiology is
putatively unrelated to leukocyte gene expression, such as neurological disorders or type 2
diabetes. Also unlike DNA sequence, methylation at some CpG sites is potentially dynamic, such
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
that disease progression can induce epigenetic changes (25, 26). To avoid such reverse causality,
identifying variants with long-term temporal stability is advantageous for EWAS.
Accordingly, ideal genomic regions to target in population epigenetic studies would show
substantial interindividual variation in DNA methylation that is both systemic (i.e. not cell type
specific) and stable over time. To identify such regions, we previously conducted deep unbiased
DNA methylation profiling in three embryonic germ layer lineages of each of ten human donors,
identifying ~10,000 correlated regions of systemic interindividual epigenetic variation (CoRSIVs)
(27). A subsequent study of 188 individuals (28) validated systemic interindividual variation (SIV)
at CoRSIVs and also identified over 70 times more genetic influence on methylation (methylation
quantitative trait loci – mQTL) than had been detected by other studies, mostly based on the
Illumina platform. Just as for detection of mQTL, interindividual variation in DNA methylation is
essential for detection of associations with disease and environmental exposures. We therefore
postulated that, relative to most probes on the Illumina array platform, those within CoRSIVs are
more likely to be reported in EWAS. Here, we describe an automated text-mining analysis testing
this hypothesis.
Across 2,203 published EWAS, Illumina probe ID occurrences link to biology
Medical Subject Headings (MeSH) comprise a defined, hierarchically organized vocabulary used
in indexing articles in PubMed. Based on disease categories in a previous study (28), and the MeSH
hierarchy under “diseases” [MeSH], we selected 11 disease categories for analysis: cancer,
cardiovascular, digestive, endocrine, hematological, immune, metabolic, neurological (including
both nervous system and mental disorders), obesity, respiratory, and urogenital (Fig. 1A). From
the full-text and supplementary tables of 2,203 publications under these 11 categories (Fig. 1A, B,
table S1), we extracted 234,586 unique HM450 / EPIC probe IDs (which we refer to as ‘probes’).
Across each of 11 disease categories, the number of publications ranged from 85 (hematological)
to 1,085 (cancer) (Fig. 1C). We chose not to use existing EWAS databases such as the EWAS Atlas
(29) or EWAS catalog (30) because these manually curated databases are inherently limited in
scope. The largest, EWAS Atlas, includes only ~1,000 studies, fewer than half as many as we
evaluated (fig. S1).
Our analysis is predicated on the assumption that probes reported in the literature in the context of
different MESH terms reflect, to some extent, associations with pathophysiology of specific
diseases. To test this, taking cancer as an example, we observed that as we increase the number of
papers in which a specific probe must be reported, the number of probes meeting the threshold
decays exponentially (Fig. 2A). Considering the 18,132 probes reported in at least one paper in the
context of cancer (but not in other disease categories), we explored the associated genes by gene-
set enrichment analysis using DisGeNET. Although the top 10 disease terms (by adjusted P value)
included several cancer-related terms, these were barely significant (Fig. 2B). When we restricted
the analysis to the 13,113 probes reported in two or more studies in the context of cancer, the top
ten disease terms were highly significant and all related to cancer (Fig. 2B). This indicates that
probes reported in at least two papers in the context of the same disease are more reliably linked
to pathophysiology.
Only 41,653 probes were reported in at least 2 papers within the same disease category (table S2).
Unsupervised hierarchical clustering of these (Fig. 2C) identified both distinct clusters of probes
uniquely associated with specific categories of disease (such as cancer and neurological disorders)
and extensive overlap in some related categories such as metabolic and obesity. Notably, excluding
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
the 13,113 probes reported only in the context of cancer, only 28,460 probes were reported in at
least two papers in the context of any of the other diseases. Similar to the results for cancer (Fig.
2B), gene-set enrichment analysis considering genes associated with probes uniquely reported in
two or more studies in the context of neurological, immune, or cardiovascular disorders detected
highly significant disease terms primarily associated with the respective categories (Fig. 2D – F;
table S3).
We compiled curated systemic interindividual variation (SIV) loci from several sources including
putative metastable epialleles (31), Illumina probes identified as exhibiting epigenetic
supersimilarity or systemic interindividual variation (32), and the 9,926 human CoRSIVs (27).
Although some of these loci were identified prior to introduction of the CoRSIV terminology, we
consider them all CoRSIVs. Of these 10,388 regions, only 1,607 (15.5%) are covered by at least
one probe on the Illumina HM450 or EPIC array (table S4). Within these 1,607 regions are 3,517
probes, which we term CoRSIV probes (average 2.2 probes/CoRSIV); table S5 annotates these
relative to proximal genes.
We next asked whether CoRSIVs reported in the literature tend to be associated with specific genic
regions. Compared to all 1,607 CoRSIVs covered on the Illumina array, those reported in at least
2 papers in the context of one of the 11 disease categories were about twice as likely to be within
200 bp of a transcription start site (TSS200) of a gene (chi-square P < 10-10) and half as likely to
be intergenic (chi-square P = 0.006) (Fig. 2G). Together, these results support the plausibility of
our analysis by indicating that occurrences of probe IDs in the literature associate with disease-
specific pathophysiology and reflect the established role of promoter methylation in transcriptional
regulation (33).
CoRSIV probes are overrepresented in most disease categories
Considering the relatively high CpG density of CoRSIVs (≥5 CpGs per 200 bp), the full set of
Illumina probes is not an appropriate background set against which to evaluate enrichment of
CoRSIV probes. We therefore modeled expected probe occurrences based on 10 sets of control
regions; within each of the 10 sets, a control region was matched to each CoRSIV based on a range
of parameters including chromosome, region size, and number of overlapping Illumina probes (see
Methods). This yielded 35,170 control probes (table S6) within 16,070 control regions (table S7).
We averaged across the ten sets of control regions to establish a stable baseline from which to
evaluate enrichment. We noted that, for most classes of disease, as we increased the threshold for
the number of papers in which a specific probe must be reported (i.e., increasing reliability), the
percentage of those probes overlapping CoRSIVs increased markedly (Fig. 3A – F); no such
increase was observed among the control probes (gray lines).
Within each disease category, we assessed enrichment of CoRSIV relative to control probes by
calculating the ratio of the weighted sum of CoRSIV probe occurrences to the average weighted
sum of control probe occurrences, with weights based on the number of publications reporting
each probe:
!"#$%&'(") *+,-./012#,3(1,%%4##("%(5
+,")#,612#,3(1,%%4##("%(5 * 1
7
$1 8 1 "!
2!
"
!#$
7
$1 8 1 '!9:;
2!
"
!#$
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
where k is the highest number of papers that each report at least 10 identical probe IDs, ni is the
number of CoRSIV probes reported in exactly i papers, pi is the total number of probes reported
in exactly i papers, and mi is the number of control probes reported in exactly i papers (across 10
control sets). We then determined statistical significance by permutation testing (Fig. 3G and fig.
S2). An annotated example of an enrichment ratio calculation (fig. S3) shows that high enrichment
ratios are largely driven by CoRSIV probes reported in a high number of papers.
We found statistically significant enrichment of CoRSIV probes in 7 of the 11 categories: cancer,
endocrine, immune, metabolic, neurological, obesity, and urogenital (Fig. 3A – 3F; fig. S4; table
S8). Results for all categories are summarized in Fig. 3H. The strongest enrichment ratios were
found within the endocrine (53.5-fold), neurological (25.7-fold), and metabolic categories (23.7-
fold); cancer and immune disorders yielded over 10-fold enrichments. Remarkably, we observed
these strong overrepresentations of CoRSIV probes despite only 15.5% of known CoRSIVs being
represented on the Illumina platform. Hence, to estimate the power advantage of profiling all
known CoRSIVs compared to using the Illumina platform, these enrichments can be multiplied by
6.5 (i.e. 1/0.155). EWAS profiling all known CoRSIVs would therefore be expected to have 345-,
166-, and 153-fold greater power to detect associations with endocrine, neurological, and
metabolic disorders, respectively (fig. S5; table S8). Even for diseases showing modest enrichment,
like cancer, urogenital disorders and obesity, CoRSIV-focused EWAS promise remarkable 70-, 45-
and 37-fold increases in power, respectively.
High-resolution analyses provide insights into category-specific enrichments
Each of the MeSH categories we evaluated comprises dozens of different diseases and disorders.
For several categories with ample numbers of papers, therefore, we utilized the MeSH hierarchy
to conduct enrichment analyses at higher resolution. For example, within the neurological category
(Fig. 4A; table S9), CoRSIVs were particularly enriched in the context of neurodevelopmental
disorders (34.6-fold; P = 1.1 x 10-187; Fig. 4B), including intellectual disability (22.5-fold; P = 5.3
x 10-71; Fig. 4C), attention deficit disorder with hyperactivity (54.7-fold; P < 2.2 x 10-308; Fig. 4D),
autism spectrum disorder (21.1-fold; P = 4.0 x 10-108; fig. S6A), and schizophrenia (12.7-fold; P =
1.0 x 10-97; fig. S6B). (Although not listed within “neurodevelopmental disorders” in the MeSH
hierarchy, many consider schizophrenia a neurodevelopmental disorder (34).) These enrichments
have not been multiplied by the 6.5 scaling factor. Remarkably, over half of the papers within the
neurodevelopmental disorders and ADHD categories, and two-thirds of those under intellectual
disability report at least one CoRSIV probe. In striking contrast, and despite comparable numbers
of studies, neurodegenerative disorders including dementia, tauopathies, and Alzheimer’s disease
showed no enrichment (Fig. 4A). In addition to neurodevelopmental disorders, our analysis (Fig.
4A) detected an enrichment of CoRSIV probes in studies associated with nervous system diseases
including nervous system neoplasms (22.3-fold; P < 2.2 x 10-308; fig. S6C) and multiple sclerosis
(23.5-fold; P = 8.4 x 10-291; fig. S6D).
Within our text-mining approach we were unable to develop a reliable automated method to focus
only on statistically significant probe occurrences. In the context of this sub-category analysis,
therefore, we tested whether restricting the analysis to statistically significant probes yields
comparable results. We manually curated all 15 papers annotated to the “Attention Deficit Disorder
with Hyperactivity” MeSH Term and tabulated nominally significant probes associated with
ADHD or environmental exposures leading to ADHD. Whereas 1,555 probe instances were
common to both approaches (fig. S7A), 1,399 were unique to the manual approach, as the
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
automated approach was blind to tables not included in the PMC Open Access Subset. Of the 1,054
probe instances unique to the automated approach, ~60% were excluded from the manual approach
because they were not directly relevant to ADHD outcomes and ~40% were because they were not
statistically significant. Overall, the manual approach, yielded an enrichment ratio of 34.6 (table
S10; P = 3.0 x 10-276; fig. S7B), comparable to the 54.7 ratio obtained by the automated approach.
Enrichment analysis within the metabolic MeSH category (fig. S8A; table S11) indicated that its
enrichment was largely driven by glucose metabolism disorders (45.7-fold enrichment; P = 7.8 x
10-170; fig. S8B), specifically type 2 diabetes (15.8-fold; P = 2.0 x 10-32; fig. S8C) and gestational
diabetes (9.3-fold; P = 8.0 x 10-27; fig. S8D), as well as lipid metabolism disorders (18.1-fold; P <
<2.2 x 10-308). With the caveat that there were a relatively small number of studies, no enrichment
was observed for type 1 diabetes. Exploring the endocrine MeSH category (fig. S9; table S12)
indicated that, like metabolism, its exceptionally strong enrichment was primarily driven by type
2 diabetes and gestational diabetes.
Lastly, breaking down cancer, the MeSH category with the largest number of papers (fig. S10A;
table S13), indicated a high degree of heterogeneity. Whereas cancer showed an 11.2-fold over-
representation CoRSIV probes (Fig. 3A), most cancer subtypes showed no enrichment. The most
striking exception is the 58.1-fold enrichment for prostate cancer (P = 1.2 x 10-87; fig. S10B);
nervous system neoplasms (22.3-fold; P < 2.2 x 10-308; fig. S10C) and neuroendocrine tumors (5.4-
fold; P = 2.7 x 10-14; fig. S10D) also showed enrichment. MeSH hierarchies and diseases showing
significant CoRSIV enrichment in the context of immune and urogenital studies are shown in fig.
S11 and table S14, S15. Additional details on studies contributing to CoRSIV enrichment at the
specific disease level are provided in tables S16 – S21.
Stable and Systemic Interindividual Variation Explains the Advantage of CoRSIVs
Wishing to gain insights into why CoRSIV-overlapping probes are so strongly over-represented in
the EWAS literature, we analyzed data on long-term stability of DNA methylation in peripheral
blood of 92 individuals each profiled twice using the Illumina HM450 methylation array, across
an average interval of six years (35). We quantified stability of probe-level methylation using the
intraclass correlation coefficient (ICC), and interindividual variation using the 2nd to the 98th
percentile of interindividual range (IIR2-98%) (36) (table S22) (see Methods). Probes with an ICC >
0.5 are considered moderately stable (35). Evaluating the 2,111 CoRSIV probes and 412,269 non-
CoRSIV probes with data available on both ICC and IIR2-98% (Fig. 5A) indicated that long-term
stability was strongly associated with interindividual range. CoRSIV probes generally exhibited
substantial interindividual variation (median IRR2-98% = 0.22), much greater than that of non-
CoRSIV probes (median IRR2-98% = 0.06). CoRSIV probes were also highly stable (median ICC =
0.82), whereas long-term stability of non-CoRSIV probes was generally poor (median ICC = 0.16).
These dramatic differences between CoRSIV and non-CoRSIV probes were also observed among
probes associated with different classes of disease (fig. S12).
In sensitivity analyses, within each MeSH category we asked whether interindividual range and
stability of DNA methylation are related to the strength of the associations, based on the numbers
of papers reporting a specific probe. For CoRSIV probes, being reported in more papers was
associated with increasing interindividual range (Fig. 5A), particularly in the cancer, neurological,
and urogenital categories (P = 0.001, 1.2 x 10-4 and 0.003 respectively) and nearly so for endocrine
(P = 0.05). For non-CoRSIV probes, on the other hand, no probe subsets showed median
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
interindividual range greater than 0.2; rather, being reported in more papers was most strongly
associated with increasing stability of methylation, most obviously within the endocrine and
neurological categories (Fig. 5A). These results indicate that interindividual variation that is
substantial and stable increases the likelihood that a probe will be reported in the literature.
Next, to test whether the systemic nature of interindividual variation at CoRSIVs is also factor we
queried the BECon database (37), which provides probe-level data on correlations of DNA
methylation between blood and four different brain regions. The median blood-brain correlation
(BECon score) was substantially higher for CoRSIV than non-CoRSIV probes, not only within the
neurological category (0.40 and 0.03 respectively) (Fig. 5B) but also within the others (fig. S13A).
A sensitivity analysis within the neurological category showed that, for both CoRSIV and non-
CoRSIV probes, being reported in more papers was associated with increasing brain-blood
concordance (Fig. 5C). In the other disease categories, significant associations for CoRSIV probes
were not detected (fig. S13B). This is consistent with the blood-brain basis of the BECon score.
Together, these analyses indicate that the over-representation of CoRSIV probes in the EWAS
literature is attributable to the substantial, systemic, and stable interindividual variation in DNA
methylation at the targeted CpG sites.
Discussion
In this study, we mined text from 2,203 published studies using the Illumina methylation array
platform, finding that the 3,517 Illumina methylation array probes overlapping CoRSIVs are
reported in the literature much more than would be expected by chance. Of 11 disease categories
analyzed, 7 showed statistically significant enrichment. Moreover, sensitivity analyses indicated
that, relative to other probes on the Illumina methylation platform, the considerable power
advantage of CoRSIV CpGs is related to their interindividual variation, which is substantial,
systemic, and stable.
We are not aware of any previous studies that have taken a similar approach to gain insights into
the thousands of EWAS that have been reported in the last 12 years using the Illumina methylation
platform. Previous studies (38, 39) utilized text-mining approaches to study disease-gene
associations in the published literature but, to our knowledge, none has been reported in the context
of DNA methylation. Additionally, previous text-mining studies have primarily focused on
constructing databases of disease-gene relationships (40). In general, Illumina probe IDs reported
in EWAS are identified based on statistical tests and can be considered as distinct data elements.
Our study establishes a precedent for utilizing probe occurrences in the literature to assess the
enrichment of specific genomic regions in EWAS, providing a novel framework for integrating
EWAS results across multiple studies. The validity of this approach is supported by our gene-set
enrichment analyses. Simply by tabulating probes reported in two or more publications within each
of several disease categories and evaluating the associated genes, we detected highly significant
enrichments for relevant disease terms.
It is interesting to consider why some classes of disease are strongly associated with CoRSIVs,
and others not. We detected the strongest enrichments for CoRSIVs in the context of neurological
disorders and metabolic disease (particularly type 2 diabetes). The enrichment for neurological
disorders is not surprising, given that nearly half of CoRSIV genes are associated with nervous
system diseases or mental disorders (27). The strong association with metabolic disorders is also
consistent with our previous findings (28); analysis of the GWAS catalog showed that, among
several different classes of disease, single nucleotide variants associated with metabolic disease
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
are most strongly enriched for mQTL at CoRSIVs. The enrichment observed in the metabolic
category was primarily driven by type 2 diabetes and gestational diabetes, with no enrichment
under type 1 diabetes. This finding aligns with the established role of gestational diabetes as a risk
factor for subsequent type 2 diabetes (41), whereas type 1 diabetes follows a distinct etiology.
Inspection of the results in the context of neurological disorders yielded the important insight that
the enrichments are limited to neurodevelopmental and not neurodegenerative disorders. This
aligns with our previous finding that the developmental establishment of methylation at CoRSIVs
occurs in the preimplantation embryo, under the influence of both genetic variation (28) and
periconceptional environment (27, 31, 42-44). The association of CoRSIVs with
neurodevelopmental disorders could be interpreted in two ways. First, considering their systemic
nature, interindividual variation in methylation at CoRSIVs could influence brain development,
directly contributing to the risk of neurodevelopmental disorders. Alternatively, associations of
CoRSIV methylation with neurodevelopmental disorders may arise if CoRSIV methylation is a
biomarker for environmental conditions during prenatal development that affect both epigenetic
development in the early embryo and subsequent neurodevelopment. Interindividual variation in
CoRSIV methylation may even represent an evolutionary mechanism for promoting
neurodiversity and adaptability in response to changing cultural and environmental demands (6,
45, 46), conferring population-level benefits while increasing individual risk for
neurodevelopmental disorders.
An epigenetic basis for neurodevelopmental disorders and type 2 diabetes is consistent with the
developmental origins paradigm, which suggests that during critical ontogenic periods,
environmental factors such as nutrition alter developmental pathways, thus affecting lifelong
susceptibility to a wide range of chronic diseases (8, 9). Extensive evidence from humans and
animal models indicates that the risks of neurodevelopmental disorders and type 2 diabetes are
strongly influenced by environment during development. The most compelling human data derive
from long-term follow up studies of individuals who were exposed to famine during early
development. For example, multiple studies in independent cohorts show that early gestational
exposure to famine is associated with a doubling of risk for schizophrenia (47, 48). A recent
analysis of survivors of the Ukraine famine of 1932-33 found that exposure to famine during early
gestation likewise results in a two-fold increased risk for type 2 diabetes in adulthood (49). Notably,
in both examples, the period of susceptibility (or critical window) during early gestation is
consistent with the early embryonic period when CoRSIV methylation is established.
On the other hand, except for prostatic and nervous system neoplasms, despite being the disease
most studied from an epigenetic perspective, CoRSIV probes were generally not over-represented
in cancer. There are several possible reasons for this. First, unlike other classes of disease, for
which investigators are often limited to profiling DNA methylation in blood, many cancer studies
compare DNA methylation in tumor and adjacent normal tissue. In this context, the systemic nature
of variation at CoRSIVs offers no advantage; likewise for studies profiling blood methylation in
the context of hematological disorders. Moreover, the original design of the Illumina array
prioritized CpG sites with aberrant methylation in tumors, often a consequence rather than a cause
of tumorigenesis. Nonetheless, prospective studies assessing DNA methylation in peripheral blood
have shown that baseline measurements of CoRSIV methylation in peripheral blood are associated
with the risk of several types of cancer (32, 50).
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
Strengths of our study include the extremely large and systematically curated data set we exploited,
comprising thousands of published studies and hundreds of thousands of data points (probe
occurrences), and our novel approach of analyzing occurrences of Illumina probe IDs to gain
biological insights. These factors enabled the detection of strong and highly significant disease-
specific associations, even for some disease categories comprising relatively small numbers of
papers. Also, of the diseases found to show significant enrichment for CoRSIVs, the ~100-fold
power advantage we estimated relative to studies using the Illumina platform is consistent with the
72-fold power advantage we reported earlier in the context of detecting mQTL (28). Further, our
sensitivity analyses using independent data sets provided an explanation for the outsized power
advantage of CoRSIVs relative to the other probes on the Illumina platform.
Nonetheless, our study is not without limitations. Despite initial efforts to employ large language
models, our automated approach was unable to reliably determine the tissue analyzed in each study.
Although most EWAS are based on blood, some profile DNA methylation within a tissue more
relevant to the disease of interest, in which case the systemic nature of CoRSIVs is not an
advantage. Cancer is one example of this, as described above. Whereas our analysis was conducted
at the probe level, many EWAS identify differentially methylated regions (DMRs) comprising
multiple CpG sites operating in a coordinated fashion. Due to difficulties determining the specific
genome build used and inconsistencies in coordinate formatting, however, we were unable to
develop an automated approach to assess overlap of DMRs with CoRSIVs. This is unfortunate
because, as CoRSIVs are themselves coordinated regions of variable methylation, DMR-level
analyses would be expected to show even stronger enrichment for CoRSIVs than probe-level
analyses. Our analysis focused exclusively on EWAS conducted using the Illumina platform.
However, CoRSIVs have been identified as top hits in studies utilizing other profiling approaches,
such as whole-genome bisulfite sequencing in the context of autism spectrum disorder (51, 52)
and reduced-representation bisulfite sequencing in the offspring of obese mothers (53). Our study
is based on the rationale that reliability of a specific probe being associated with a disease is based
on reporting frequency, rather than P values determined in each study. We were unable to develop
a reliable automated approach for linking P values to probe occurrences. Our manually curated
analysis focusing on nominally significant probes reported in the context of ADHD, however,
yielded a highly significant enrichment comparable to that obtained using the automated approach,
supporting the validity of our results. Nonetheless, the ability to analyze additional information
from each study, such as sample sizes and P values, would likely be advantageous.
Conclusion
Our findings have important implications for the fields of epigenetic epidemiology and population
epigenetics. The finding that neurodevelopmental disorders are enriched for CoRSIV-overlapping
probes has implications for determining underlying mechanisms of disorders such as schizophrenia,
which have thus far been elusive. To accelerate progress in understanding epigenetic epidemiology
of human disease, it will be crucial for the field to either augment the Illumina platform to improve
coverage of CpG sites within CoRSIVs or adopt a next-generation standardized platform for
profiling CoRSIV methylation in humans. Our results suggest that for many of the thousands of
EWAS that have employed the Illumina platform it will be hugely advantageous to utilize the same
DNA samples to perform methylation profiling at CoRSIVs. This presents an outstanding
opportunity to rapidly advance our understanding of epigenetic etiology of disease. Lastly, the
current list of known CoRSIVs is incomplete. The largest published screen for CoRSIVs (27) was
based on only ten White American GTEx donors. To broadly explore associations between
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
interindividual variation and risk of disease, it will be crucial to map out CoRSIVs across diverse
human ancestry groups and environmental contexts.
References and Notes:
1. P. M. Visscher et al., 10 Years of GWAS Discovery: Biology, Function, and Translation.
Am J Hum Genet 101, 5-22 (2017).
2. V. Tam et al., Benefits and limitations of genome-wide association studies. Nat Rev Genet
20, 467-484 (2019).
3. L. A. Hindorff et al., Potential etiologic and functional implications of genome-wide
association loci for human diseases and traits. Proc Natl Acad Sci U S A 106, 9362-9367
(2009).
4. L. Bomba, K. Walter, N. Soranzo, The impact of rare and low-frequency genetic variants
in common disease. Genome Biol 18, 77 (2017).
5. G. M. Cooper, J. Shendure, Needles in stacks of needles: finding disease-causal variants in
a wealth of genomic data. Nat Rev Genet 12, 628-640 (2011).
6. J. M. McClellan et al., An evolutionary perspective on complex neuropsychiatric disease.
Neuron 112, 7-24 (2024).
7. E. E. Eichler et al., Missing heritability and strategies for finding the underlying causes of
complex disease. Nat Rev Genet 11, 446-450 (2010).
8. R. A. Waterland, C. Garza, Potential mechanisms of metabolic imprinting that lead to
chronic disease. Am.J.Clin.Nutr. 69, 179-197 (1999).
9. R. A. Waterland, K. B. Michels, Epigenetic epidemiology of the developmental origins
hypothesis. Annu Rev Nutr 27, 363-388 (2007).
10. E. Birney, G. D. Smith, J. M. Greally, Epigenome-wide Association Studies and the
Interpretation of Disease -Omics. PLoS Genet 12, e1006105 (2016).
11. V. K. Rakyan, T. A. Down, D. J. Balding, S. Beck, Epigenome-wide association studies for
common human diseases. Nat Rev Genet 12, 529-541 (2011).
12. R. F. Hillary et al., Blood-based epigenome-wide analyses of 19 common disease states: A
longitudinal, population-based linked cohort study of 18,413 Scottish individuals. PLoS
Med 20, e1004247 (2023).
13. S. L. Schaffner, M. S. Kobor, DNA methylation as a mediator of genetic and environmental
influences on Parkinson's disease susceptibility: Impacts of alpha-Synuclein, physical
activity, and pesticide exposure on the epigenome. Front Genet 13, 971298 (2022).
14. N. Gladish, S. M. Merrill, M. S. Kobor, Childhood Trauma and Epigenetics: State of the
Science and Future. Curr Environ Health Rep 9, 661-672 (2022).
15. A. L. Non, Social epigenomics: are we at an impasse? Epigenomics 13, 1747-1759 (2021).
16. C. International HapMap, The International HapMap Project. Nature 426, 789-796 (2003).
17. C. Genomes Project et al., A global reference for human genetic variation. Nature 526, 68-
74 (2015).
18. R. Pidsley et al., Critical evaluation of the Illumina MethylationEPIC BeadChip microarray
for whole-genome DNA methylation profiling. Genome Biol 17, 208 (2016).
19. M. Bibikova et al., High density DNA methylation array with single CpG site resolution.
Genomics 98, 288-295 (2011).
20. M. Bose et al., Evaluation of microarray-based DNA methylation measurement using
technical replicates: the Atherosclerosis Risk In Communities (ARIC) Study. BMC
Bioinformatics 15, 312 (2014).
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
21. Z. Xu, J. A. Taylor, Reliability of DNA methylation measures using Illumina methylation
BeadChip. Epigenetics 16, 495-502 (2021).
22. Y. B. Shvetsov et al., Intraindividual variation and short-term temporal trend in DNA
methylation of human blood. Cancer Epidemiol Biomarkers Prev 24, 490-497 (2015).
23. I. Zaimi et al., Variation in DNA methylation of human blood over a 1-year period using
the Illumina MethylationEPIC array. Epigenetics 13, 1056-1071 (2018).
24. E. Hannon et al., Assessing the co-variability of DNA methylation across peripheral cells
and tissues: Implications for the interpretation of findings in epigenetic epidemiology.
PLoS Genet 17, e1009443 (2021).
25. J. I. Young et al., Genome-wide brain DNA methylation analysis suggests epigenetic
reprogramming in Parkinson disease. Neurol Genet 5, e342 (2019).
26. J. Viana et al., Schizophrenia-associated methylomic variation: molecular signatures of
disease and polygenic risk burden across multiple brain regions. Hum Mol Genet 26, 210-
225 (2017).
27. C. J. Gunasekara et al., A genomic atlas of systemic interindividual epigenetic variation in
humans. Genome Biol 20, 105 (2019).
28. C. J. Gunasekara et al., Systemic interindividual epigenetic variation in humans is
associated with transposable elements and under strong genetic control. Genome Biol 24,
2 (2023).
29. M. Li et al., EWAS Atlas: a curated knowledgebase of epigenome-wide association studies.
Nucleic Acids Res 47, D983-D988 (2019).
30. T. Battram et al., The EWAS Catalog: a database of epigenome-wide association studies.
Wellcome Open Res 7, 41 (2022).
31. M. J. Silver et al., Independent genomewide screens identify the tumor suppressor
VTRNA2-1 as a human epiallele responsive to periconceptional environment. Genome
Biol 16, 118 (2015).
32. T. E. Van Baak et al., Epigenetic supersimilarity of monozygotic twin pairs. Genome Biol
19, 2 (2018).
33. P. A. Jones, Functions of DNA methylation: islands, start sites, gene bodies and beyond.
Nat Rev Genet 13, 484-492 (2012).
34. M. J. Owen, M. C. O'Donovan, A. Thapar, N. Craddock, Neurodevelopmental hypothesis
of schizophrenia. Br J Psychiatry 198, 173-175 (2011).
35. J. M. Flanagan et al., Temporal stability and determinants of white blood cell DNA
methylation in the breakthrough generations study. Cancer Epidemiol Biomarkers Prev 24,
221-229 (2015).
36. C. J. Gunasekara et al., A machine learning case-control classifier for schizophrenia based
on DNA methylation in blood. Transl Psychiatry 11, 412 (2021).
37. R. D. Edgar, M. J. Jones, M. J. Meaney, G. Turecki, M. S. Kobor, BECon: a tool for
interpreting DNA methylation findings from blood in the context of brain. Transl
Psychiatry 7, e1187 (2017).
38. J. Kim, J. J. Kim, H. Lee, An analysis of disease-gene relationship from Medline abstracts
by DigSee. Sci Rep 7, 40154 (2017).
39. J. L. Liu, M. Zhao, A PubMed-wide study of endometriosis. Genomics 108, 151-157 (2016).
40. V. Kuleshov et al., A machine-compiled database of genome-wide association studies. Nat
Commun 10, 3341 (2019).
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
41. M. V. Diaz-Santana, K. M. O'Brien, Y. M. Park, D. P. Sandler, C. R. Weinberg, Persistence
of Risk for Type 2 Diabetes After Gestational Diabetes Mellitus. Diabetes Care 45, 864-
870 (2022).
42. N. J. Kessler, R. A. Waterland, A. M. Prentice, M. J. Silver, Establishment of
environmentally sensitive DNA methylation states in the very early human embryo. Sci
Adv 4, eaat2624 (2018).
43. M. S. Estill et al., Assisted reproductive technology alters deoxyribonucleic acid
methylation profiles in bloodspots of newborn infants. Fertil Steril 106, 629-639 e610
(2016).
44. R. A. Waterland et al., Season of conception in rural gambia affects DNA methylation at
putative human metastable epialleles. PLoS Genet 6, e1001252 (2010).
45. J. K. Burns, Reconciling 'the new epidemiology' with an evolutionary genetic basis for
schizophrenia. Med Hypotheses 72, 353-358 (2009).
46. A. P. Moczek et al., The role of developmental plasticity in evolutionary innovation. Proc
Biol Sci 278, 2705-2713 (2011).
47. D. St Clair et al., Rates of adult schizophrenia following prenatal exposure to the Chinese
famine of 1959-1961. JAMA 294, 557-562 (2005).
48. E. Susser et al., Schizophrenia after prenatal famine. Further evidence. Arch Gen
Psychiatry 53, 25-31 (1996).
49. L. H. Lumey, C. Li, M. Khalangot, N. Levchuk, O. Wolowyna, Fetal exposure to the
Ukraine famine of 1932-1933 and adult type 2 diabetes mellitus. Science 385, 667-671
(2024).
50. A. Ghantous et al., Epigenome-wide analysis across the development span of pediatric
acute lymphoblastic leukemia: backtracking to birth. Mol Cancer 23, 238 (2024).
51. Y. Zhu et al., Placental methylome reveals a 22q13.33 brain regulatory gene locus
associated with autism. Genome Biol 23, 46 (2022).
52. Y. Zhu et al., Placental DNA methylation levels at CYP2E1 and IRS2 are associated with
child outcome in a prospective autism study. Hum Mol Genet 28, 2659-2674 (2019).
53. A. Sasaki, K. E. Murphy, L. Briollais, P. O. McGowan, S. G. Matthews, DNA methylation
profiles in the blood of newborn term infants born to mothers with obesity. PLoS One 17,
e0267946 (2022).
54. J. P. T. Higgins, S. Green, C. Cochrane, Cochrane handbook for systematic reviews of
interventions. (Cochrane Collaboration, Oxford, 2022).
55. S. Wei et al., Ten Years of EWAS. Adv Sci (Weinh) 8, e2100727 (2021).
56. J. Pinero et al., DisGeNET: a comprehensive platform integrating information on human
disease-associated genes and variants. Nucleic Acids Res 45, D833-D839 (2017).
57. E. Y. Chen et al., Enrichr: interactive and collaborative HTML5 gene list enrichment
analysis tool. BMC Bioinformatics 14, 128 (2013).
58. T. K. Koo, M. Y. Li, A Guideline of Selecting and Reporting Intraclass Correlation
Coefficients for Reliability Research. J Chiropr Med 15, 155-163 (2016).
Acknowledgments:
Funding: The authors acknowledged support from the following grants:
National Institutes of Health / National Institute of Diabetes and Digestive and Kidney
Diseases grant R01DK125562 (RAW)
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
National Institutes of Health / National Institute of Diabetes and Digestive and Kidney
Diseases grant R01DK129265 (RAW)
United States Department of Agriculture grant CRIS 3092-51000-065-003S (RAW)
Author contributions:
Conceptualization: RAW
Methodology: WC, RAW
Data curation: WC
Formal analysis: WC, UM, AY
Funding acquisition: RAW
Supervision: CJG, RAW
Writing – original draft: WC, RAW
Writing – review & editing: all authors
Competing interests: Authors declare that they have no competing interests.
Data and materials availability: All data are available in the main text or the supplementary
materials. All codes and pipelines to analyze data and generate figures are available on
GitHub (https://github.com/computational-epigenetics-section/CoRSIV-text-mining).
Supplementary Materials
Materials and Methods
Figs. S1 to S15
Tables S1 to S22
References (54-58)
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
Fig. 1. Workflow for mining Illumina probe IDs from the EWAS literature. (A) Hierarchical
tree of Medical Subject Headings (MeSH) terms included in the analysis. (B) Workflow for
paper selection and probe extraction process. (C) Number of studies included within each disease
category. Category labels are abbreviations of the MeSH terms in panel A, except for
“Neurological”, which combines MeSH terms “Mental Disorders” and “Nervous System
Diseases”.
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
Fig. 2. Probe ID occurrences in the literature map to biology. (A) The decay plot for cancer
illustrates that as we increase the number of papers in which a specific probe must be reported,
the number of probes meeting the threshold decays exponentially. (B) Gene set enrichment
analysis (GSEA) of genes annotated to the 18,132 probes cited in at least one cancer-related
paper (and unique to cancer). The top ten terms (by adjusted P value) are moderately significant
and include some unrelated to cancer. Restricting the analysis to the 13,113 probes reported in
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
two or more cancer-related studies yields highly significant terms, all related to cancer. (Terms
are shaded as in Fig. 1C.) (C) Only 41,653 probes were reported in two or more papers within at
least one of the 11 disease categories. Unsupervised hierarchical clustering of these reveals
distinct clusters of probes associated with cancer and neurological disorders, and overlap in
related categories such as metabolic and obesity. (D-F) GSEA of probes reported in at least two
papers within each of the (D) neurological, (E) immune, and (F) cardiovascular categories
identifies highly significant terms primarily linked to the respective classes of disease. (G)
Compared to all CoRSIVs covered on the Illumina platform (dotted lines), those reported in at
least 2 papers within each of several disease categories tend to locate near the transcription start
site (TSS) of genes (P < 10-10) and not in intergenic regions (P = 0.006). Each colored bar
represents CoRSIVs reported in ≥ 2 papers within each disease category; color-coding follows
Figure 1C. Dashed line represents all 1,607 CoRSIVs covered on the Illumina platform.
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
Fig. 3. In most disease categories, probes reported in the literature are enriched for overlap
with CoRSIVs. (A-F) Stacked decay-enrichment plots for the cancer, endocrine, immune,
metabolic, neurological, and urogenital categories, respectively. As individual probes are
required to be reported in more papers, the number of probes meeting the threshold decays (top
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
panels) while the percent overlap with CoRSIVs increases (lower panels). Insets (red boxes)
report the enrichment ratio and number of CoRSIV probes and papers driving each enrichment.
(G) Results of permutation testing (100,000 iterations) for the cancer category. (H) Summary of
results across all 11 categories.
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
Fig. 4. CoRSIVs are associated with neurodevelopmental but not neurodegenerative
disorders. (A) Term-specific results for the MeSH hierarchy encompassing the neurological
category. Statistically significant enrichments are indicated by blue shading. Numbers indicate
the total number of papers analyzed within each MeSH term. Only categories showing (i)
statistically significant enrichment, and (ii) including at least one CoRSIV probe reported in at
least two papers are colored. (B-D) Stacked decay and enrichment plots for neurodevelopmental
disorders, intellectual disability, and attention deficit disorder with hyperactivity, respectively.
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
Fig. 5. Stable, systemic interindividual variation appears to explain CoRSIVs’ advantage.
(A) Six-year test-retest stability data based on the HM450 platform (35) show that long-term
stability (intraclass correlation coefficient, ICC) is dependent on interindividual range (IIR2-98%,
see Methods). The median ICC of the 2,111 CoRSIV-overlapping probes (filled star) is much
higher than that of the 412,269 non-CoRSIV probes (hollow star). Category-specific sensitivity
analyses (shading) show that, for CoRSIV probes, being reported in more papers is associated
with increasing interindividual range (rightward trend indicated by arrow). Only data points
representing at least 15 probes are shown. (B) Analysis of probe-level brain vs. blood correlation
data (BECon) (37) shows that, of those reported in at least two papers within the neurological
category, CoRSIV probes (blue) show higher brain-blood correlation than non-CoRSIV probes
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
(grey). Dotted lines indicate median values. (C) Sensitivity analysis; being reported in more
neurological papers is associated with increasing BECon score, for both CoRSIV and non-
CoRSIV probes. Only data points representing at least 15 probes are shown.
. CC-BY 4.0 International licenseIt is made available under a
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted February 12, 2025. ; https://doi.org/10.1101/2025.02.10.25322021doi: medRxiv preprint
