Keywords
Cerebellar ataxia; Child neglect; Epilepsy; Epstein-Barr Virus (EBV); Exoneration; Failure-to-thrive; Heterozygosity; Mitochondrial diseases; Nutrient deprivation; POLG-related disorders; Seizures; Viral encephalitis; Wrongful incarceration.
Abstract
A proper defense against alleged child neglect can be challenging. Accurate convictions by prosecutorial teams are especially important, as most exonerations do not include child neglect conflation. Common signs and symptoms of child neglect include failure-to thrive, cerebellar ataxia, seizures, and epilepsy. Here we highlight the intersection of various etiological agents capable of eliciting clinical manifestations that mirror child neglect. Case studies further illustrate how recurrent infections by viral agents of encephalitis, layered over host genetics in the form of heterozygosity, can reveal a viral-induced haploinsufficiency condition that is easily conflated with child neglect. To help prevent such conflations in the future, we systematically analyzed over 1.1 million alleles across ~2,300 genetic loci that are at greatest risk for child neglect conflation cases. Our analytical pipeline, comprising multivariate statistical analyses and Topological Data Analysis (TDA), highlights each of 1.1 million allelic contributions towards child neglect conflation. We advise that future child neglect cases involve establishing the absence of recurrent viral agents of encephalitis as well as ensuring the absence of any null mutation at each of the ~2,300 genetic loci at greatest risk for child neglect conflation. Caution should thus be exercised when approaching child neglect cases involving heterozygosity at any of these genetic loci, especially when host immunological evidence suggestive of recurrent viral infections is present in the pediatric setting.
INTRODUCTION
Child neglect by nutrient deprivation or starvation commonly results in Failure-To-Thrive (FTT) and involuntary convulsions yielding pediatric epilepsy and seizures [1]. However, these clinical manifestations are often associated with numerous viral agents of encephalitis as well as certain, well-documented genetic disorders, including mitochondrial diseases [2,3]. In the case of recurrent viral infection, pediatric FTT often manifests during the course of viral infection, pathophysiology, and overall damage to afflicted host tissues [2]. Pediatric growth during recurrent viral infections is thus often attenuated as the host immune system resolves the infectious agent, requiring host resource diversion away from metabolism and growth activities [2]. Similarly, inborn errors of metabolism inherited through the germline from parental DNA to the child can result in deficits in the child’s ability to convert dietary nutrients into proper metabolic products [4]. In the case of such genetic disorders, oftentimes mitochondrial activities are greatly attenuated, and the clinical manifestations of such mitochondrial diseases additionally can mirror viral encephalitis cases as well as true, bona fide child neglect by nutrient deprivation [3]. Additionally, most exonerations do not involve child neglect cases. The need to prosecute only bona fide cases of child neglect, and to prevent accidental conflation with viral etiologies or host genetic disorders, is thus remarkably high as those who are wrongfully incarcerated for alleged child neglect charges are often without exoneration recourse.
METHODS
Data accessibility
The National Registry of Exonerations (NRE) is maintained as a collaborative endeavor across three major law schools: The University of California - Irvine Newkirk Center for Science and Society, the University of Michigan Law School, and the Michigan State University College of Law. The NRE database encapsulates over 3,000 exonerations in the United States since 1989. Our data cutoff for the NRE database was on September 28, 2023. Custom Python scripts and data string manipulations were performed to data wrangle each exoneration case to map all NRE case specific tags per the NRE’s public coding manual [5-7].
Genetic loci that contribute to clinical manifestations of Failure To-Thrive (FTT) and/or epilepsy/seizure were collated using several databases and datamining approaches. For epilepsy and seizure phenotypes (MedGen UID: 4506; Concept ID: C0014544), we accessed the National Center for Biotechnology Information’s (NCBI) Genetic Testing Registry (GTR) accession for GeneDx (GTR orgid: 26957), a genetic testing company located in Gaithersburg, MD, USA. Using NCBI GTR, we extracted the full 1,501 gene list from the GeneDx EpiXpanded Panel (GTR test accession: GTR000569723.1) which is routinely used as a genetic testing and clinical diagnostic service for epilepsy patients [8,9]. To enrich for FTT manifestations, a locus estimate was generated for POLG like genes by extracting a list of 1,136 nuclear-encoded mitochondrial gene products documented by the MitoCarta 3.0 project [10-12]. Both the GeneDx gene list (Supplementary File 1) and the MitoCarta 3.0 database (Supplementary File 2) were then collated into a master locus list by removing duplicate genes resulting in a final list comprising 2,368 unique genetic loci (Supplementary File 3).
To systematically retrieve all allelic information clinically known for each of the 2,368 unique genes, we wrote custom web-scraping Python scripts leveraging the following libraries: requests (v2.31.0), bs4 (v0.0.1), html5lib (v1.1), numpy (1.26.1), pandas (2.1.1), soupsieve (v2.5), and urllib3 (v2.0.7) [13-17]. Subsequent assembly of scraped data was organized and retained using pandas (v2.2.2) module alongside AWK scripts to collate all 2,368 scrape operations into a single master dataframe [18]. All Python (v3.9.18) code was executed in a Windows Subsystem for Linux 2 (WSL2) environment running a command line interface (CLI) of Ubuntu (v22.04.4 LTS). Custom AWK scripts were executed as shell scripts in BASH CLI via threaded WSL2 [18]. Thus, a total of 1,135,851 alleles across 2,368 unique genes, derived from the collation of GeneDx and MitoCarta 3.0 data, was initially scraped from the NCBI ClinVar portal [13-17]. Allelic data were last scraped from NCBI ClinVar on May 17, 2024.
Economic data regarding the costs for the sequencing of the human genome were obtained from the National Institute of Health’s (NIH) National Human Genome Research Institute (NHGRI), last accessed on December 26, 2023 [19,20]. The utilized NHGRI dataset tracked sequencing costs across two decades, from September 30, 2001 to May 31, 2022.
Data Analytics and visualization
For each of the ~1.1 million ClinVar-scraped alleles across 2,368 genes, raw allele counts regarding each of six NCBI ClinVar clinical classifications [i.e., pathogenic (P), likely pathogenic (LP), conflicting classifications (CC), uncertain significance (US), likely benign (LB), and benign (B)] were collated using pandas dataframes in Python (Supplementary File 4) [13-17]. Relative allelic clinical categorical frequencies were then calculated per each genetic locus (Supplementary File 5) [13-17]. A total of nine genes were removed due to zero values in all six ClinVar clinical classification categories, resulting in a total unique gene list of 2,359 loci carrying a total of 1,135,851 documented alleles [13-17].
To examine the phenotypic classification similarities across genetic loci, all normalized ClinVar clinical categorical frequencies for the ~1.1 million alleles across 2,359 genes were mapped in six-dimensional space, where each axis represented one of the six NCBI ClinVar categorical allele frequencies (i.e, pathogenic, likely pathogenic, conflicting classifications, uncertain significance, likely benign, and benign). Thus, each gene is represented by a single point in six-dimensional space, where its location in that space is given by the gene’s specific NCBI ClinVar categorical allele frequency distribution. Dimensional reduction via Topological Data Analysis (TDA) was performed using the Kepler Mapper module (v2.0.1), with deprecated numpy (v1.26.1) and scikit (v1.3.2) libraries, in a Python (v3.9.18) environment encapsulated by a threaded WSL2/Ubuntu (v22.04.4 LTS) [21]. The computational pipeline first calculated the Euclidean distance between each pair of genes, yielding a distance matrix for the complete set of genes (i.e., n = 2,359 genes). This distance matrix was next used as input for kmapper (v2.0.1) with the following parameters: centrality as a filter function, a cover of ten bins with 25% overlap, and scikit-learn’s (v1.3.2) DBSCAN with epsilon set to 1 as a clusterer [21]. The resulting cluster definitions containing each cluster’s gene members can be found in supplementary file 6 and an interactive TDA visualization of all 2,359 loci in HTML format is available as supplementary file 7. Multivariate statistical analyses were performed using custom Python (v3.9.18) scripts in containerized virtual environments comprising WSL2 and Ubuntu v22.04.4 (Supplementary File 8). Principal component analysis (PCA) utilized the scikit-learn (v1.5.1) library for StandardScaler, PCA, and Pipeline class methods (Supplementary File 8B,C) [22-24]. Linear regression to generate correlation coefficients was completed by native pandas (v2.2.2) methods (Supplementary File 8D). For additional dimensional reduction and visualization, t-distributed stochastic neighbor embedding (t-SNE) analysis was performed via scikit-learn (v1.5.1), with a learning rate set to 50 (Supplementary File 8E) [25-27]. Partitional clustering of n observations (i.e., categorical allele counts) into k clusters was performed by first removing all but the first two principal components (i.e., PC1 and PC2), followed by iterative k-means clustering operations to arrive at 10 defined clusters (Supplementary File 8F) [28 31].
Data-wrangling was performed using either Python/WSL2/Ubuntu via the pandas (v2.2.2) module or via Microsoft (MS) Excel 365. Pairplots, scatterplots, barplots, heatmaps, and pie charts were either generated programmatically using Python libraries including matplotlib (v3.9.0) and seaborn (0.13.2), or via MS Excel 365. BioRender was utilized for finalizing all art assets. Circos [32] is a useful tool to visualize relational data in a circular layout, with algorithmic implementations in both R as Circlize [33] and Python as pyCirclize [34]. We employed a Circos visualization of our TDA output using pyCirclize (v1.6.0) in a Python environment (v3.9.18) running on a threaded WSL2 Ubuntu (v22.04.4 LTS) installation [34].
RESULTS
Exoneration challenges in child neglect cases
Since 1989, there have been over 3,385 exonerations in the United States accounting collectively for 40,488 years lost due to wrongful imprisonment. The National Registry of Exonerations tracks each exoneree’s case with detailed coding tags, and data-mining these tags reveals that roughly a tenth of all 3,385 exonerations involve cases centered around children in some fashion (Figure 1).
Figure 1: Exoneration cases in the United States involving children constitute a minimal fraction of total exonerations.Data regarding exonerations in the U.S were sourced from the National Registry of Exonerations (NRE),which was last accessed on September 28, 2023. A total of 3,385 exonerees were found in the NRE database at that time,and the case-specific metadata tags were datamined for relative frequencies across all cases.A tenth of all exoneration cases from 1989 to Sep 2023 involved tags related to children in some fashion.
As expected, most exonerations revolve around capital crimes involving homicide and murder, and thus most exoneration efforts are not focused on pediatric related contestations,including child neglect conflation cases (Figure 1) [5-7].
Conflation risk between viral infection and child neglect
Infectious viral agents in pediatric cases are known drivers of not only viral encephalitis but also a cadre of clinical manifestations that approximate child neglect, including Failure-To-Thrive (FTT), involuntary convulsions, and seizures [2]. Such viral agents include many herpesviruses such as herpes simplex virus-1 (HSV-1) and HSV-2, varicella-zoster virus (VZV or chickenpox), Human Herpesvirus 4 (HHV-4 or Epstein-Barr Virus or EBV), Human Herpesvirus 5 (HHV-5, or cytomegalovirus or CMV), and Human Herpesvirus 6 (HHV-6 or roseolovirus) (Table 1) [2]. Viral infections in neonates can exert severe host pathophysiology costs that often result in stunted growth, which can resemble failure-to-thrive and thus be easily conflated with bona-fide child neglect via purposeful nutrient deprivation [2]. In contrast, starvation of a newborn (i.e, bona fide child abuse) can result in epileptic seizures of varying grades, which can also be easily conflated with the involuntary convulsions that typify viral encephalitis. Much precedent exists in clinical case studies that have documented recurrent, and oftentimes, sub-lethal, self-limiting viral infections in numerous pediatric cases that present similarly to cases of child neglect [35-46].
Case studies involving recurrent viral encephalitis
In 1984, a 2-month-old female baby was admitted due to involuntary convulsions accompanying fever (Figure 2A) [35]. Clinicians identified an enlarged forehead (i.e., fontanelle), slight anemia, and a blood serum EBV IgG (i.e., immunoglobulin type G) that indicated successful affinity maturation, suggestive of recurrent EBV infection. Additionally, Cerebrospinal Fluid (CSF) samples of the baby indicated high titers of EBV IgG, indicating viral encephalitis. In fact, the baby would be hospitalized a total of four times within an approximately six month period, with each visit recapitulating nearly identical clinical manifestations. After the fourth hospitalization for seizures, the pediatric patient would not exhibit any involuntary convulsions while clinicians tracked her across the next six years of her early childhood. By 1993, the patient was reported to have fully recovered. Thus, a low-grade but persistent EBV infection drove epilepsy and seizures in a neonate in 1984 who had presented with fever (i.e., febrile), involuntary convulsions, anterior fontanelle enlargement, and anemia, a list of manifestations that can be easily conflated with bona-fide child neglect by nutrient deprivation (Figure 2A) [35].
Likewise, a 2-month-old male baby (Baby M) was admitted to a Pediatric Intensive Care Unit (PICU) in 2019 (Figure 2B). Upon admission to the PICU, Baby M presented febrile, with involuntary convulsions, an enlarged forehead, and slight anemia, nearly mirroring the 1984 case (Figure 2) [35].
Figure 2: Recurrent Epstein-Barr Virus (EBV) in pediatric cases drives viral meningoencephalitis manifestations. Sub-lethal and self-limiting signs and symptoms indicative of persistent, low-grade EBV infections manifest in stereotyped patterns.(A):In 1984, a female baby was hospitalized four times from 2 months to 7.5 months of age, and each hospitalization involved a combination of fever, involuntary convulsions, enlarged forehead (i.e., fontanelle),and a low-grade anemia defined by low levels of hemoglobin (g/dL).Blood serum and Cerebrospinal Fluid (CSF) revealed elevated levels of high-affinity EBV IgG, which indicates recurrent EBV infection.(B):In 2019, a male baby (aka Baby M) was hospitalized at 2 months of age for nearly identical manifestations of fever, involuntary convulsions, enlarged forehead, and anemia defined by low levels of hemoglobin (g/dL). A spinal tap was not performed on Baby M but blood panels did reveal elevated levels of affinity-mature serum EBV IgG, indicative of recurrent EBV infection and suggestive of EBV-induced viral encephalitis. Additionally, Baby M presented with other clinical manifestations such as pallor, metabolic lactic acidosis, and acute liver failure. Many of these additional manifestations are seen in patients diagnosed with POLG-Related Disorders (PRD).Baby M harbors a single POLG lesion (POLG S933R) and is thus heterozygous (i.e., exhibiting POLGrelated disorder trait or PRDT).Combined with recurrent EBV infections, a convincing argument can be made for viral-induced haploinsufficiency at the POLG locus as the underlying etiology for Baby M’s clinical manifestations.
However, Baby M did exhibit additional clinical manifestations, including cerebellar ataxia, acute liver failure, metabolic lactic acidosis, and a gram-negative (E. coli) sepsis. While under PICU care, high antibody titers of EBV IgG were detected in Baby M’s blood samples, indicative that Baby M experienced recurrent or persistent EBV infection. A spinal tap was not performed for Baby M, and thus Cerebrospinal Fluid (CSF) samples were not generated. Taken together, alongside neonatal manifestations of epilepsy and seizures, it is highly likely that Baby M experienced recurrent EBV-induced viral encephalitis. It is unknown as to whether Baby M continued to experience febrile convulsions for the next six months of his life as child protective services separated Baby M’s biological mother from Baby M within 2-3 months of Baby M’s PICU visit. Beyond Baby M (Figure 2B) and the 1984 clinical case study [35], the investigation of EBV infection in neonates and early childhood has been well-documented in the clinical literature for nearly half a century [47- 54].
Medical genetics teams at the PICU also performed extensive sequencing of DNA extracted from Baby M’s blood samples to determine if Baby M harbored any defective alleles in genes associated with epilepsy and seizure [8,9]. Samples collected from Baby M during his PICU stay permitted over 100 genetic loci to be screened via Next-Generation Sequencing (NGS) methodologies, and Baby M was determined to be heterozygous for the POLG locus, specifically carrying a POLG S933R missense substitution allele. Baby M did not otherwise exhibit any other genetic defects.
The POLG gene, mitochondria, and risk for child neglect conflation
The POLG locus comprises a nuclear-encoded mitochondrial DNA polymerase subunit (Figure 3) [55].
Figure 3: Overview of mitochondrial function with an emphasis on POLG functions. (A):Each human cell contains up to 1,000 mitochondrial organelles, and each single mitochondrion harbors 2-10 copies of its circular, mitochondrial chromosome.(B):On a single mitochondrial chromosome,the mitochondrial DNA (mtDNA) comprises a total of 37 essential genes which encode 13 mRNAs, 22 tRNAs and 2 rRNAs.(C): All 13 mRNAs encode essential proteins for oxidative phosphorylation (OXPHOS) which is how we produce the bulk of our ATP molecules necessary for life.(D-F):All mitochondrial organelles still execute mtDNA replication, transcription, and translation within the mitochondrial matrix.(D):mtDNA replication requires the nuclear-encoded POLG1 and POLG2 loci.The mtDNA polymerase complex is a trimeric complex that includes one POLG1 (aka POLG) subunit harboring exonuclease and polymerase activities and two POLG2 subunits as scaffold.(E):Transcription of all 37 mitochondrial genes relies on the nuclear-encoded POLRMT gene.(F):Translation of all 13 mtDNA-encoded mRNAs requires mitochondrial ribosome operation within the mitochondrial matrix.
Mitochondria are organelles that transform dietary nutrients (e.g. ingested glucose) into high-energy Adenosine Triphosphate (ATP) molecules, 90% of which are synthesized via a process known as Oxidative Phosphorylation (OXPHOS) (Figure 3C) [56,57]. OXPHOS occurs on the innermost mitochondrial membrane known as the cristae membrane which taps into a membrane potential that powers ATP formation in the mitochondrial matrix (Figure 3C) [56,57]. Each human cell contains up to a thousand mitochondrial organelles, and each mitochondrial organelle harbors 2-10 copies of the circular, mitochondrial chromosome (Figures 3A,B) [58-60]. There are a total of 37 mitochondrial genes that reside on such mitochondrial DNA (mtDNA), and 13 encode proteins that are essential for OXPHOS activity while the remaining 24 mtDNA genes service mitochondrial ribosomes in the mitochondrial matrix (Figure 3A) [61-63]. However, none of the 37 mtDNA genes encode any aspect of DNA polymerase activity. Instead, across endosymbiotic time, such genes have relocated to the human nucleus, and include POLG1 (aka POLG) found on chromosome 15 and POLG2 located on chromosome 17 (Figure 3D) [64-66]. POLG encodes a bifunctional enzyme, with a polymerase domain for catalyzing phosphodiester bond formation and an exonuclease domain for proofreading daughter strand fidelity (Figure 3D) [55]. Each locus (i.e., POLG and POLG2) is individually expressed, and their gene products assemble within the mitochondrial matrix as a trimeric mtDNA polymerase complex, composed of one catalytic POLG subunit and two scaffolding POLG2 subunits (Figure 3D) [55,67-69]. Mitochondrial disease patients present across a wide spectrum, often in pediatric cases accompanied by manifestations such as failure-to-thrive, cerebellar ataxia, seizures, and epilepsy [70-74]. POLG-Related Disorders (PRD) encapsulate such mitochondrial disease patients and traditionally include POLG homozygous recessive Loss Of-Function (LOF) conditions [4,75]. However, POLG heterozygotes are not completely normal, and when manifesting clinically, POLG heterozygotes draw from a nearly-identical phenotypic list, including but not limited to FTT, cerebellar ataxia, and epilepsy [76]. In the case of Baby M’s clinical manifestations, POLG heterozygosity alone can explain numerous phenotypic presentations, including FTT, cerebellar ataxia, seizures, and epilepsy as such manifestations have been routinely documented for pediatric POLG heterozygotes in nearly two decades of peer-reviewed clinical literature [76]. However, layering recurrent EBV infections on top of Baby M’s POLG heterozygosity state likely amplified his clinical manifestations in overall grade and severity. This intersection between host genetics (i.e., POLG) and host experience (i.e., recurrent EBV infections) likely culminates phenotypically as a viral-induced POLG haploinsufficiency induction event in Baby M’s specific instance. More importantly, Baby M’s manifestations highly resembled that of child abuse via nutrient deprivation. Consequently, Baby M’s legal guardian was incarcerated for child neglect by nutrient deprivation with a legal defense that failed to discuss viral-induced haploinsufficiency mimicry of child neglect manifestations.
Survey of at-risk loci for child neglect conflation cases
Mitochondrial diseases in pediatric cases, characterized by failure-to thrive manifestations, are not solely driven by mutations at the POLG locus [3,4,77-80]. Modern machine-learning approaches, such as MitoCarta 3.0, have cataloged over 1,136 nuclear-encoded mitochondrial gene products [10-12]. Likewise, known genes underlying epilepsy and seizures are well documented and comprise in-depth genetic testing targets for aberrant neurological manifestations in the clinical setting [8,9]. GeneDx offers one such testing panel, which includes 1,501 known genes associated with epilepsy and seizures [8,9]. We combined both MitoCarta 3.0 and GeneDx gene lists, removing all duplicate loci, into a single master gene list (Supplementary Files 1-3). The collated set of genes (Supplementary File 3) thus constitutes genetic loci (n = 2,368 genes) that are enriched for either Failure-To-Thrive (FTT), developmental delay, epilepsy, or seizure phenotypes, all of which are clinical manifestations of not only host genetic deficiencies or recurrent viral infections but also in child neglect cases involving nutrient deprivation.
NCBI ClinVar is a database that tracks clinical variants or mutant alleles pertaining to each human gene. Clinical assessments are regularly updated for each clinical variant’s entry, and a clinical classification is provided for any particular mutant allele. NCBI ClinVar provides clinical classifications for each allele as one of six clinical categories: pathogenic, likely pathogenic, conflicting classifications, uncertain significance, benign and likely benign. For genetic loci that contribute to FTT, developmental delay, epilepsy or seizures, the last two ClinVar allele classifiers (i.e., benign and likely benign) run the least risk of conflation with bona fide child neglect [13-17]. However, the other four categories, especially pathogenic and likely pathogenic, are strong phenotypic drivers and thus can account for clinical manifestations. We programmatically scraped all 2,368 gene accessions for allelic classifiers from NCBI ClinVar as these loci might contribute to conflation with bona fide child neglect cases (Figure 4A).Over a million alleles (n = 1,135,851 alleles) across all 2,368 loci were analyzed, with an average of ~480 alleles per locus ranging from 0 known alleles up to a maximum of 14,417 alleles at the NF1 locus (Figure 4A). Log2 histogram analysis of total allele numbers per gene reveals that all 2,368 loci exhibit a normal distribution in ten-percentile bin increments (Figure 4B). The POLG gene occupies the upper tail end of this allelic abundance distribution (Figure 4B). We next sorted all 2,368 genes by the sum of all known alleles that might contribute to child neglect conflation risk by summating all alleles except for benign or likely benign categorical classifiers (Figure 5A).
When viewed by total alleles at greatest risk for conflation, the POLG gene is found in the top ~1% of the gene list, specifically the 28th of 2,368 genes (Figure 5A). Analysis solely by raw allele counts per locus favors those genes with the greatest number of alleles. To detect categorical allelic patterns agnostic of total known alleles, we next calculated the relative clinical classification frequencies per genetic locus, and re-sorted by the sum of all normalized allele category frequencies while excluding benign and likely benign categories since these two ClinVar categories are least likely to contribute towards child neglect conflation cases. When sorted by normalized categorical frequencies, the POLG gene is found midway in the list at position 1,472 out of 2,368 genes (Figure 5B). The top 1% of genes, visualized after performing a deprecating, normalized categorical allele frequency sort, is dominated by pathogenic clinical classifiers, likely suggesting that such loci are functionally essential and do not tolerate diverse sets of mutations (Figure 5B). Additionally, most of these loci, when viewed under decrementing, normalized ClinVar categorical allele frequencies, exhibit exceptionally low total allele depositions in the ClinVar database (Supplementary Files 4,5). Multivariate statistical analyses revealed that all six ClinVar allele categories exhibited noteworthy variance differentials, as measured qualitatively via pairplots (Supplementary File 8A) or quantitatively using principal component analysis (Supplementary Files 8B,C), linear regression and correlation coefficient analysis (Supplementary File 8D), t-SNE (Supplementary File 8E), or k-means clustering (Supplementary File 8F). Roughly 91.5% of the variance in the ~1.1 million allelic data regarding NCBI ClinVar clinical classification categories can be explained by just three principal components (Supplementary Files 8B,C). Most importantly, both t-SNE and k-means clustering of principal components regarding the ~1.1 million categorical allele counts strongly suggested a rich dataset with robust potential for cluster identification.
TDA of at-risk loci for child neglect conflation cases
While analysis of total allele counts (Figure 4) or relative clinical classifier frequencies (Figure 5) provides convenient rankings of each genetic locus, a major drawback to this approach is that ranked lists do not reveal phenotypic similarities or differences across all genes.
Figure 4: Total allele numbers drawn from a list of ~2,300 genes that either drive neurological manifestations (i.e., seizures and/or epilepsy) or are known to sustain key mitochondrial activities. Custom Python scripts were written to scrape the NIH’s NCBI ClinVar database for all known alleles across 2,368 genetic loci as of May 17, 2024. (A): Genetic loci are collectively visualized with a log2 transformation of scraped allele numbers. When sorted by total allele numbers in descending order, POLG is the 35th of 2,368 total genes with the greatest number of alleles. POLG is indicated by the yellow star, and the Pareto line for gene coverage is shown as red line. (B): Histogram analyses in 10% bin increments of total allele number per genetic locus illustrate that most of the 2,368 genes follow a binomial distribution centered at the 60% percentile bin. POLG (yellow star) is found in the second to last bin.
Figure 5: Gene priority lists are sorted based on conflation risk visualizing total allele number or normalized categorical allele frequencies. For each allele across 2,368 scraped genes, the precise NCBI ClinVar clinical classification category was tracked as: Pathogenic (P), Likely Pathogenic (LP), Conflicting Classifications (CC), Uncertain Significance (US), Likely Benign (LB), and Benign (B). (A): Total allele sums across all clinical categories, except for likely benign and benign, were calculated and visualized in a descending sort (P + LP + CC + US). Shown here are the top thirty genes of which POLG (yellow star) is the 28th gene of then 2,368 genes when sorted by total alleles per genetic locus that are at-risk for child neglect conflation. (B): All alleles for each genetic locus were then normalized to reveal the ClinVar clinical categorical allele frequency distributions. Descending sorts were then performed on the sums of the relative frequencies for normalized sums of Pathogenic (P), Likely Pathogenic (LP), Conflicting Classifications (CC), and Uncertain Significance (US). Shown here are the top thirty genes of which POLG is the 1,472nd of 2,368 genes when sorted by relative ClinVar clinical allele categorical frequencies.
To examine additional phenotypic patterns, we employed Topological Data Analysis (TDA) after representing each gene as a single point in six-dimensional space, where each of six axes represents one of the six NCBI ClinVar clinical classification frequencies (i.e., pathogenic, likely pathogenic, conflicting classifications, uncertain significance, benign and likely benign). After removing nine genetic loci for lacking any known alleles in the NCBI ClinVar database, we performed TDA on a list of 2,359 genes given by their normalized ClinVar phenotypic classification categories. TDA categorized the 2,359 genes across a total of ten clusters, whose mapper graph had the shape of a flare (Figure 6).
Figure 6: Topological Data Analysis (TDA) of genes at risk for child neglect conflation reveals ten distinct clusters in a flare configuration.Each of 2,359 genes was represented in six-dimensional space using the gene’s relative NCBI ClinVar categorical allele frequency values.Custom Python scripts leveraging Kepler Mapper were then used to perform Topological Data Analysis (TDA) on all 2,359 genes to examine proximal and distal relationships.A total of 10 TDA clusters were identified in a topological flare configuration.The POLG locus (yellow star) occupies both cluster 0 (c0) and cluster 1 (c1), which are clusters defined by average ClinVar allelic categories dominated by uncertain significance.Cluster 9 (c9), in contrast, occupies the other terminus of the TDA flare,comprising of only six genes, and is phenotypically characterized by exceptionally high frequencies of just the ClinVar pathogenic category.The number of genes found within each TDA cluster is indicated in brown.
The POLG gene was found within two adjacent clusters; POLG is one of 935 genes found in cluster 0 (c0) and also one of the 959 genes comprising cluster 1 (c1) (Figure 6). Of note, c0 and c1 are dominated by genes that exhibit high allele frequencies in the uncertain significance allele category while the distally positioned, terminal clusters c8 and c9 primarily comprise genes with exceptionally high pathogenicity allele classifications (Figure 6). Visualization by Circos plots of the TDA clusters additionally illustrates the flare configuration of all ten clusters (Figure 7).
Figure 7: Circos plot visualization of normalized, categorical allele frequencies after TDA. Custom Python scripts leveraging pyCirclize were applied to Kepler Mapper’s TDA output using frequency-normalized categorical allele data as input. A total of ten TDA clusters (c0-c9) were identified in a topological flare configuration.The diagram generated by pyCirclize used a single track, with links between identical genes.The POLG locus (yellow star) is located within the first two clusters (i.e., c0 and c1). To avoid visual clutter, only every 20th gene is displayed within each cluster. Cluster membership by each gene indicates phenotypic similarity, as computed by NCBI ClinVar clinical classification data.
Taken together, genes found in clusters c0 and c1 are POLG-like in their ClinVar relative allelic categorization frequencies, and thus might also contribute to child neglect conflation cases in a fashion similar to that of the POLG locus (Figures 6,7). For further loci evaluation, the reader is invited to examine a full gene membership list of all TDA clusters (Supplementary File 6) alongside an interactive flare of all ten TDA clusters (Supplementary File 7).
DISCUSSION
In pediatric cases, Failure-To-Thrive (FTT) and developmental delay are both clinical manifestations with broad underlying etiologies [1,4,40]. Likewise, neurological deficits that include involuntary convulsions, epilepsy, and seizures collectively draw from similar root causes, namely host genetics, viral infection history, and/or nutrient deprivation (Figure 8).
Figure 8: Summary of major etiologies driving pediatric Failure-To-Thrive (FTT) manifestations.Failure-to-thrive as a clinical manifestation is an overly broad phenotypic category with many etiological culprits.(A): In true child neglect cases, insufficient or inadequate nutrients supplied to a baby can result in FTT. (B): In pediatric cases involving persistent or recurrent viral infections, particularly with viral agents that can cause encephalitis, viral pathophysiology and host damage can result in pediatric FTT. (C): Host genetics can also drive pediatric FTT.Certain in-born errors of metabolism,including POLG-Related Disorders (PRD) that manifest with Mitochondrial DNA Depletion Syndromes (MDDS), can prevent growth due to inadequate ATP production or abnormal metabolism. (D): Viral-induced haploinsufficiency is possible in cases involving host heterozygosity. For example, Baby M is a POLG heterozygote (+/S933R) who experienced recurrent EBV infections. This viral exposure / host genotype combination revealed an EBV-induced haploinsufficiency condition in which Baby M’s sole Wildtype (WT) copy of POLG was insufficient to maintain normal mtDNA replication needs, presumably due to mitochondrial toxicity contraindications. Viral-induced haploinsufficiency can thus result in pediatric FTT. (E):The Venn overlay (zones I, II, and III) between any combinatorial permutation of nutrient deprivation, viral agent exposure, and host genetic disorders are areas that are at the highest risk of child neglect conflation cases. (F): A simplified Venn diagram indicates the complexity involved in all three driving factors for pediatric FTT, including nutrient deprivation, viral infection history, and host genetic deficiencies.
For all three etiologies, only nutrient deprivation by withholding nutrients from a neonate qualifies for bona-fide child neglect (Figure 8A). Any evidence of recurrent viral infectious agents known to cause viral encephalitis (e.g., Baby M’s elevated serum levels of affinity mature EBV IgG) must be ruled out as an explanation for FTT, epilepsy or seizures (Figure 8B). Likewise, any inherited defects due to inborn errors of metabolism inherited via germline transmission (e.g., Baby M’s inherited POLG S933R allele) must be carefully considered as these conditions can also drive clinical manifestations such as cerebellar ataxia, FTT, developmental delay, epilepsy, or seizures (Figure 8C). In the case of Baby M, POLG heterozygosity must additionally be seriously considered as a major etiological component towards clinical manifestations (Figure 8D) as an extensive body of clinical literature illustrates that POLG heterozygotes exhibit numerous phenotypes, particularly for pediatric epilepsy patients [76]. Further, recurrent EBV infections in neonates alone are known to cause epilepsy and seizures [35-46]. Meanwhile, neurological deficits are seen in a wide spectrum of clinical manifestations that includes epilepsy and seizures for mitochondrial disease patients diagnosed with POLG-related disorders [4,75]. Taken together, Baby M’s POLG heterozygosity likely potentiates a vulnerability for enhanced severity in clinical manifestations involving EBV infections (Figures 2B,8D), best explained by EBV-induced POLG haploinsufficiency wherein Baby M’s one normal (wild-type, WT) POLG allele is insufficient to sustain mtDNA replication needs during the course of viral pathophysiology. In developing neonates, cerebellar neurons exhibit elevated POLG requirements and lack of sufficient mtDNA replication drives cerebellar ataxia as a clinical manifestation [4,75]. Mitochondrial diseases cause OXPHOS dysfunction and reduce ATP production in afflicted host cells, and lack of ATP production will manifest clinically as if the child is starving (i.e., FTT, seizure, or epilepsy). POLG is not the sole gene driving mitochondrial disease diagnoses, and this study provides a conservative estimate (n = 2,368) of the number of genes that might pose a risk for accidental child neglect conflation by analyzing genes associated with epilepsy and seizures or gene products that exhibit a localization pattern to the mitochondria (Supplementary File 3).
The greatest conflation risk exists at the intersection of all three etiological drivers underlying FTT manifestations in pediatric cases (Figures 8E,9).
Figure 9: Conflation risk of child neglect cases with either viral encephalitis and/or inborn errors of metabolism, including POLG-Related Disorders (PRD).Each of three factors, such as nutrient starvation, past viral infection history, or host genetic disorders, can drive a common pediatric manifestation list that includes Failure-To-Thrive (FTT),Developmental Delay (DD), epilepsy, seizures, abnormal MRIs and EEGs, and liver abnormalities (i.e., hepatopathy or acute liver failure).Zone II (i.e., the intersection of all three etiological drivers) is a zone of greatest conflation risk as it becomes difficult to tease apart the underlying contributions of nutritional status, viral exposure, and/or host genetics towards a shared clinical manifestation list.
The challenge when constructing a prosecutorial case centered on bona-fide child neglect charges is to thus rule out any recurrent viral agents of encephalitis in the pediatric patient’s history alongside targeted DNA sequencing efforts to reveal WT alleles at each of 2,368 genetic loci that are at greatest risk for conflation. As Baby M’s case illustrates, even POLG heterozygosity, and thus harboring just a single mutant POLG allele, potentiates risk for etiological conflation (Figure 8D) [76]. Although sequencing each of the 2,368 genetic loci per child in alleged child neglect cases might sound intractable, the deprecating costs of Next-Generation Sequencing (NGS) technologies makes the endeavor economically feasible and deployable at-scale across numerous cases (Figure 10).
Figure 10: Deprecating costs of Next-Generation Sequencing (NGS) technologies permits cost-effective surveillance of each of the ~2,300 genes at risk for child neglect conflation. Impressively, human genome sequencing costs in the last twenty years have decreased from ~$100 million to the hundred dollar range today. This is primarily driven by Illumina NGS technologies involving Sequencing-By-Synthesis (SBS).The ability to sequence all 2,368 genes in the child involved in child neglect cases is thus not beyond reach. Establishing Wild-Type (WT) alleles for all 2,368 genes is recommended to prevent the wrongful incarceration of a parent due to conflation of bona-fide child neglect (i.e., via nutrient deprivation) with some form of host genetic abnormality (i.e., either homozygous recessive or heterozygosity coupled to haploinsufficiency induction).Data are sourced from the NIH’s National Human Genome Research Institute (NHGRI), last accessed in Dec 2023.
For instance, the cost of sequencing a human genome was approximately $100 million USD in 2001 but today, NGS permits sequencing a single human genome at a reasonable cost of a few hundred dollars (Figure 10) [19,20,81,82].
CONCLUSIONS AND RECOMMENDATIONS
As most exonerations do not involve child neglect cases, there is thus a greater onus for successful and accurate prosecution (Figure 1) [5-7]. Any blood serum or Cerebrospinal Fluid (CSF) samples containing elevated levels of affinity-mature IgG directed against known viral encephalitis agents must therefore be ruled out as a cause of clinical manifestations. Recurrent viral infectious agents in pediatric cases are known to stunt growth (i.e., failure-to-thrive) and in viral encephalitis diagnoses, epilepsy, seizure, and involuntary convulsions are common phenotypes (Table 1) [2]. Furthermore, mitochondrial diseases such as POLG-related disorders share similar manifestations such as developmental delay, epilepsy, and seizures [4,75]. Additionally, the genetics underlying mitochondrial diseases are not always governed by complete dominance rules (76). For example, while patients harboring two defective POLG genes (aka POLG -/-) exhibit the clinical manifestations underlying POLG-related disorders, the POLG homozygous dominant (POLG +/+) genotypes exhibit a differing phenotype compared to POLG heterozygotes (POLG +/-) [76].
In fact, POLG heterozygosity coupled to recurrent EBV infection provides positive evidence to fully account for and explain the severity of Baby M’s clinical manifestations [76]. To avoid the wrongful incarceration of caregivers for alleged child neglect charges, we recommend that prosecution teams proceed with the following:
1. From the baby, obtain blood serum antibody titers and test for the absence of high-affinity IgG directed against any of the 19 viral agents of encephalitis (Table 1) [2].
2. From the baby, collect blood samples for DNA extraction and sequence all 2,368 genetic loci and test for the absence of problematic alleles (Supplementary File 3) [8-12].
We recommend that investigators proceed with a prosecution case of child neglect only if (1) the baby does not exhibit any evidence of recurrent viral encephalitis as evidenced by lack of directed IgG and (2) the baby does not harbor any defective alleles and thus exhibits a +/+ genotype at each of the 2,368 genetic loci. Of note, heterozygosity (+/-) cannot be assumed to be normal as evidenced in Baby M’s viral-induced POLG haploinsufficiency case [76].
By ruling out either a viral or host genetics cause of the pediatric clinical manifestations, we finally recommend that prosecution teams obtain direct evidence of nutrient starvation of the child by looking for broad-scale vitamin and mineral deficiencies in either blood or urine samples paired with the physiological consequences of specific vitamin and mineral deficiencies as evidenced on the body and/or organs of the afflicted pediatric patient [1].
By following these recommendations, we strongly believe that the chances for the wrongful incarceration of caregivers for alleged child neglect would be reduced. Routine DNA sequencing due to exceptional cost reductions should greatly facilitate this endeavor and numerous funding mechanisms are already in place [83,84].
REFERENCES
4. Rahman S. Mitochondrial disease in children. J Intern Med. 2020; 287: 609-633.
23. Hotelling H. Relations between two sets of variates. Biometrika. 1936; 28: 321-377.
29. Lloyd S. Least squares quantization in PCM. IEEE Trans Inform Theory. 1982; 28: 129-137.
34. Shimoyama Y. Circular visualization in python (circos plot, chord diagram). pyCirclize. 2024.
45. Millichap JG. Epstein-Barr virus neurologic complications. Pediatr Neurol Briefs. 2015; 29: 88.
64. Archibald JM. Endosymbiosis and Eukaryotic Cell Evolution. Current Biology 2015; 25: R911–R921.
74. Lim A, Thomas RH. The mitochondrial epilepsies. Eur J Paediatr Neurol. 2020: 47-52.
78. DiMauro S. Mitochondrial diseases. Biochim Biophys Acta. 2004; 1658(1-2): 80-88.
79. Schapira AH. Mitochondrial disease. Lancet. 2006; 368(9529): 70 82.
80. Ylikallio E, Suomalainen A. Mechanisms of mitochondrial diseases. Ann Med. 2012; 44: 41-59.
Citation
Sawyer ME, Rahama A, Gaetjens ZE, Total CR, Burton AD, Gacura MD, et al.(2024) The Risk of Conflating Child Neglect with Viral Induced Haploinsufficiency. Ann Appl Microbiol Biotechnol J 5: 15.