The Use of Zebrafish in Transcriptome Analysis of the Early Effects of Mutations Causing Early Onset Familial Alzheimer’s Disease and Other Inherited Neurodegenerative Conditions

Abstract

The degree to which non-human animals can be used to model Alzheimer’s disease is a contentious issue, particularly as there is still widespread disagreement regarding the pathogenesis of this neurodegenerative dementia. The currently popular transgenic models are based on artificial expression of genes mutated in early onset forms of familial Alzheimer’s disease (EOfAD). Uncertainty regarding the veracity of these models led us to focus on heterozygous, single mutations of endogenous genes (knock-in models) as these most closely resemble the genetic state of humans with EOfAD, and so incorporate the fewest assumptions regarding pathological mechanism. We have generated a number of lines of zebrafish bearing EOfAD-like and non-EOfAD-like mutations in genes equivalent to human PSEN1, PSEN2, and SORL1. To analyze the young adult brain transcriptomes of these mutants, we exploited the ability of zebrafish to produce very large families of simultaneous siblings composed of a variety of genotypes and raised in a uniform environment. This “intra-family” analysis strategy greatly reduced genetic and environmental “noise” thereby allowing detection of subtle changes in gene sets after bulk RNA sequencing of entire brains. Changes to oxidative phosphorylation were predicted for all EOfAD-like mutations in the three genes studied. Here we describe some of the analytical lessons learned in our program combining zebrafish genome editing with transcriptomics to understand the molecular pathologies of neurodegenerative disease.

Keywords

Alzheimer’s disease early onset Alzheimer’s disease gene expression profiling genetic models transcriptome zebrafish

INTRODUCTION

The question of what initiates the pathological processes leading to Alzheimer’s disease (AD) is difficult to address at the molecular level since access to tissue from living brains decades before cognitive changes is extremely limited. Therefore, detailed molecular-level analysis requires the use of appropriate disease models. An ideal model of AD would mimic the genetic state and other biology of the disease as closely as possible and enable analysis of disease progression through time. However, when mouse knock-in models of familial AD mutations were analyzed in the early 2000 s, they lacked the amyloid plaque and neurofibrillary tangle histopathology of the human disease [1 –3]. Cognitive changes were subtle and found mostly in aged mice [1, 2] (which, additionally, limits the rate of experimentation). At the time, ‘omics technologies were in their infancy and none of these knock-in mutant mice were compared to wild type (WT) mice using transcriptomic analysis.

By overexpressing mutant forms of the early onset familial Alzheimer’s disease (EOfAD) genes APP, PSEN1, and PSEN2 in transgenic mice, AD-like histopathologies with earlier cognitive changes can be obtained, and these models now overwhelmingly dominate animal model research in AD (reviewed in [4]). However, numerous studies, including transcriptome analyses, have suggested that the disease phenotypes of these models may be poorly representative of human AD [5, 6].

The App^NL - F, App^{NL - G - F} mice [7] and App^{NL - G - F} rat [8] possess App genes with humanized amyloid-β (Aβ) sequences and multiple EOfAD mutations knocked into single App alleles. Interpretation of bulk brain tissue transcriptome data from the App^{NL - G - F} mouse model is obfuscated by an apparent, statistically significant change in the proportion of microglial cells in this tissue [9], which was also observed in an independent single-cell RNAseq analysis of this model [10]. Since different cell types express genes at different levels, changes in the proportions of cell types between tissues being compared can result in detection of transcriptome differences that can be mis- (or over-) interpreted as changes in gene expression.

The refinement of genome editing technologies in the 2010 s (TALENs and then CRISPR Cas9 [11]) facilitated creation of knock-in models of human disease in mice (where embryonic stem cell technology had, in any case, allowed targeted mutagenesis since 1987 [12, 13]), and also in other animals. This has promoted analysis of genetic variants that increase the risk of the late onset form of Alzheimer’s disease (LOAD) such as the initiative MODEL-AD (Model Organism Development and Evaluation for Late-onset AD, [14]). However, the analysis of simple mouse knock-in models of EOfAD mutations has languished.

In 2014 the Alzheimer’s Disease Genetics Laboratory of the University of Adelaide began a program of generating EOfAD-like mutations in zebrafish in order to reveal the earliest cellular/molecular changes associated with the disease, primarily by exploiting ‘omics technologies. This work has revealed some unexpected advantages of using zebrafish, particularly for transcriptomic analysis.

BASIC ZEBRAFISH BIOLOGY

At first glance, zebrafish might not appear to be a suitable model for studies of neurodegenerative disease. Unlike mammals, fish have the ability to regenerate most tissues after injury, including neural tissue [15]. This likely contributes to a difficulty in inducing obvious neurodegeneration in adult zebrafish genetically [16]. However, induction by chemical means is possible (reviewed by [17]). Another occasional disadvantage of zebrafish is that, due to an ancient whole-genome duplication specific to the teleost (bony fish) lineage, approximately 1 in every 5 genes in zebrafish is duplicated. This affects the zebrafish equivalents of the AD-relevant human genes APP (appa and appb in zebrafish [18]) and APOE (apoea and apoeb in zebrafish [19]). The existence of these duplicates complicates strategies involving knock-in mutations and the breeding of particular mutant genotypes for analysis. A specific issue relevant to AD research is that zebrafish have not yet been demonstrated to produce an Aβ-equivalent peptide from either their Appa or Appb proteins. In fact, Moore et al. [20] have suggested that zebrafish do not produce such a peptide as their APP-equivalent proteins appear to lack appropriate β-secretase cleavage sites. Nevertheless, the effects of Aβ on brain biology can be studied by injecting the peptide into the brains of adult zebrafish [21, 22] and this has been exploited in brain transcriptome analysis [23].

In counterbalance to the disadvantages described above, zebrafish display numerous characteristics felicitous to AD research and approaches exploiting ‘omics technologies. We have shown that zebrafish preserve particular human presenilin gene functions that have been lost in the Muridae family of rodents that includes mice and rats. Specifically, zebrafish can produce an equivalent of the truncated “PS2V” isoform produced by HMGA1-directed alternative splicing of PSEN2 transcripts in AD brains while the Muridae have lost this ability [24, 25]. The PS2 V isoform reduces cell death under oxidative stress and can suppress the unfolded protein response while increasing γ-secretase activity [25 –27]. Interestingly, zebrafish preserve this function in their PSEN1-orthologous gene (psen1) rather than in their PSEN2 orthologue (psen2) indicating that the function arose in the ancestral presenilin gene and prior to the divergence of the teleost and tetrapod lineages 450 million years ago [25, 28].

The nearly half a billion years separating the human and zebrafish evolutionary lineages has important implications for, particularly, transcriptome analysis. Little conservation is seen in expression responses of particular genes under disease conditions in humans compared to zebrafish. However, since natural selection acts to maintain biological functions that promote survival and reproduction, the responses of sets of genes representing particular cellular functions can be expected to show a higher degree of conservation [16, 29].

A single mating event between two zebrafish can spawn well over 100 progeny that hatch after 2–3 days. Initially these progeny (“larvae”) are 4 mm long and similar in overall size to mosquito larvae with simple (but, nevertheless, vertebrate) nervous systems. If disease-related phenotypes (e.g., altered behavior) can be detected at larval stages, then the small size and free-living nature of the larvae enables them to be arrayed in microtiter plates for high-throughput screening of chemical libraries to identify therapeutically active compounds [30] (Fig. 1).

Fig. 1

Experimental workflow for behavioral screening of chemical libraries for potential therapeutic compounds using larval zebrafish. A) Zebrafish are capable of spawning over 100 embryos in a single mating event which develop rapidly, hatching within 2-3 days. B) The small size of larvae (~4 mm in length) allows them to be assayed in 96-well plates, allowing high-throughput screening of chemical libraries. An observable disease phenotype, such as behavior, is analyzed to evaluate the therapeutic effects of the different compounds (represented by different colors in the plate). C) Zebrafish heterozygous for a mutation in dnajc6 causative of juvenile-onset Parkinson’s disease are in-crossed to produce a family of both wild type and disease model larvae. At 6 days of age, their behavior can be monitored to predict their genotype using linear discriminant analysis (LDA). Linear discriminants 1 and 2 (LD1 & LD2) represent the two most discriminating behavioral features from LDA (unpublished data). Compounds that can shift the behavior of mutant individuals towards a “wild type” state (e.g., shifting the behavior of dnajc6^-/- individuals [blue dots] towards that of wild type individuals [red dots]) can be identified as potential therapeutically active compounds. (Note: some image elements are modified from those of DBLS and Servier).

Zebrafish become fertile from 3–5 months of age, although this generation time can be shortened to less than two months with intensive feeding [31]. Under laboratory conditions, the fish remain fertile for up to 2 years. Ultimately, they grow to 2–5 cm in length and can live for up to 5 years.

INTRA-FAMILY ANALYSIS REDUCES GENETIC AND ENVIRONMENTAL NOISE IN ‘OMICS ANALYSES

An advantage of the small size of zebrafish and their large family sizes is that families of siblings derived from single matings of zebrafish pairs can be raised together in uniform environments, i.e., a single tank or tanks side-by-side in a single recirculated water aquarium system. The families of siblings can be bred to contain a variety of genotypes that can be compared in ‘omics analyses. Our experience with transcriptome analysis of individual, entire zebrafish brains (of 7–20 mg wet mass) as samples is that this allows the detection of subtle differences in the expression of sets of genes that would otherwise not be seen. Figure 2 illustrates a typical intra-family brain transcriptome analysis. A fish heterozygous for a dominant EOfAD-like mutation is outbred to a WT (non-mutant) fish to generate a family of siblings half of which are heterozygous mutant and half of which are WT. The family is aged to 6 months and then all individuals simultaneously euthanized by the standard method of plunging into ice-water. This minimizes any stress due to handling (which may itself alter patterns of gene expression), euthanizes the fish rapidly, and keeps their bodies at close to 0°C while brains are removed into a RNA-stabilizing solution on ice (e.g., RNAlater®). Alternatively, the brains may be snap-frozen on dry ice for later nuclei isolation for single nucleus RNA sequencing, snRNAseq. After brain removal, the remainder of bodies are dissected to confirm the sex of individuals (as this can sometimes be difficult to distinguish with certainty visually) and to remove samples for DNA purification and PCR to determine an individual’s genotype. An advantage of this experimental structure is that, if there is no obvious physical phenotype of a mutation, then observers are blinded as to genotype until PCR-testing is completed. Interestingly, while we usually try to use equal numbers of male and female zebrafish in our analyses, analysis of transcriptomes from entire brains has not revealed strong sex-dependent differences (e.g., Fig. 2A in [32], and Fig. 3A in [33]). This contrasts with analysis of mouse brain transcriptomes where a substantial proportion of expression variability can be attributed to sex (e.g., see Fig. 3B in [9]). Recently, however, we have seen sex-dependent differences while analyzing the effects of zebrafish models of the childhood dementia, Sanfilippo syndrome/Mucopolysaccharidosis type III (MPS III). Mutations causing MPS III have more severe pathological effects than those causing EOfAD, and the sex-dependent MPS III transcriptome divergence may represent sex-dependent difference(s) in capacity to respond to this greater disruption of brain biology (data not shown).

Fig. 2

Intra-family brain transcriptome analysis in zebrafish. A) A zebrafish heterozygous for an EOfAD-like mutation is mated with a WT fish (+/+). This generates a family of progeny containing approximately equal numbers of heterozygous mutants and WT siblings. These are raised together in similar environments (e.g., the same tank) until 6 months of age. The family is then euthanized, and tissues harvested for RNAseq, for confirmation of sex, and for genotyping by polymerase chain reactions (PCR). B) n = 6 of these fish per genotype are employed in RNAseq analysis.

Fig. 3

KEGG gene sets [57] significantly altered in zebrafish brains in at least two mutants of the genes psen1 and/or psen2. Cells in the heatmap are colored by statistical significance (the negative log10 of the p-value), with the brighter colors appearing more statistically significant. The FDR-adjusted p-value is not shown if it surpassed the threshold of 0.05. The numbers inside the cells represent the harmonic mean p-values (hmp) after correction for multiple testing (a combination of raw p-values of FRY [61], CAMERA [59] and GSEA [58], described below in the section OTHER LESSONS LEARNED FROM TRANSCRIPTOME ANALYSIS IN ZEBRAFISH). This plot was generated from the gene set enrichment analysis results described in [33] for psen1 and [36] for psen2.

The intra-family analysis strategy also allows sensitive detection of differences between multiple mutant genotypes, such as fish homozygous, heterozygous, or WT for one mutation, or homozygous/heterozygous for two mutant alleles of one gene, or homozygous/heterozygous for mutations in two different genes. As an example of this advantage, we were able to investigate why reading frame-preserving dominant mutations in the presenilin genes can cause EOfAD in humans, while frameshifting dominant mutations do not, instead causing an inherited skin condition, familial acne inversa (fAI), without EOfAD [34]. We generated a family of siblings from a pair of fish heterozygous for either mutation type [33]. Specifically, the zebrafish EOfAD mutation model psen1^T428del/+ (an exact equivalent of the human PSEN1^T440del mutation [35]) was mated to a fAI mutation model, psen1^W233fs/+, to give progeny with the genotypes psen1^+/+ (WT), psen1^T428del/+ (EOfAD), psen1^W233fs/+ (fAI), or psen1^T428del^/W233fs (transheterozygous). Bulk RNAseq was performed on six individual brains of each of the WT, EOfAD, and fAI genotypes at 6 months of age, when zebrafish are recently sexually mature. Any transheterozygous fish were ignored as they do not model any known human genotype. Comparisons of brain gene expression were then performed between WT and EOfAD, and WT and fAI genotypes. This analysis, and a similar intra-family analysis of analogous mutations in the gene psen2 [36], found significant alteration of expression of the KEGG OXIDATIVE PHOSPHORYLATION (OX PHOS) gene set by EOfAD-like mutations while frameshifting mutations in these genes altered cell signaling gene sets such as those involving Notch signaling. Although subtle, the similarity in the observations made for these two different mutation types in two structurally and functionally similar genes, psen1 and psen2 (Fig. 3) increases our confidence that these effects represent real and significant differences in mutation action.

The importance of the reduction of environmental and genetic noise facilitated by intra-family analysis is illustrated by a recent, unpublished transcriptome analysis comparing the effects of homozygous recessive mutations causing MPS III in zebrafish brains at 3 months of age (Fig. 4). A pair of zebrafish, both doubly heterozygous for mutations modelling those causing MPS III subtypes MPS IIIA and MPS IIIC (loss-of-function mutations in sgsh and hgsnat respectively), were mated to produce 8 possible progeny genotypes including sgsh^+/+, hgsnat^+/+ (WT), and sgsh^-/-, hgsnat^+/+, (MPS IIIA) and sgsh^+/+, hgsnat^-/- (MPS IIIC). In parallel, another pair of zebrafish, both doubly heterozygous for mutations modelling those causing MPS III subtypes MPS IIIB and MPS IIIC (MPS IIIB is caused by recessive mutations in naglu), were mated to produce progeny genotypes including naglu^+/+, hgsnat^+/+ (WT), and naglu^-/-, hgsnat^+/+ (MPS IIIB), and naglu^+/+, hgsnat^-/- (MPS IIIC). For the MPS IIIA and MPS IIIC intra-family comparison to WT, 7 fish of each genotype were available from one family of progeny, while for the comparison of MPS IIIB and MPS IIIC to WT, limited progeny numbers forced us to use fish from three separate spawnings (two from one parental pair and one from a second parental pair). The additional genetic and environmental variability included due to the multiple spawnings and parent pairs meant that no significantly altered gene sets could be detected in the comparisons of MPS IIIB and MPS IIIC to WT, while numerous altered gene sets were detectable for the comparisons made using progeny from only one family (i.e., the intra-family comparisons of MPS IIIA and MPS IIIC to WT).

Fig. 4

The importance of experimental design in transcriptome studies. A) Schematic of the experimental design using three families of zebrafish to compare MPS IIIB and MPS IIIC to WT (+/+) zebrafish. Two parental pairs of zebrafish (P1 and P2) were used to generate 3 spawning events. The number of fish per genotype analyzed in each spawning is indicated. B) Schematic of the experimental design using one family of zebrafish to compare MPS IIIA to MPS IIIC. C) The number of statistically significant KEGG gene sets [57] we detected using FRY [61] in each MPS III subtype and experiment. Details of this experiment are available in a preprint: https://www.biorxiv.org/content/10.1101/2023.08.02.550904v1.full.

OXIDATIVE PHOSPHORYLATION IS ALTERED EARLY BY EOfAD-LIKE MUTATIONS IN psen1, psen2, AND sorl1

In the years between 2014 and 2021, we generated a number of knock-in mutation models of EOfAD-like mutations in psen1, psen2, and sorl1 [32 , 36–40]. We also generated models of non-EOfAD-like mutations in psen1 and psen2 [16 , 39]. Bulk transcriptome analysis of 6-month-old brains from all these models showed that changes in the OX PHOS gene set were common to the EOfAD-like mutations but not necessarily the non-EOfAD-like mutations [9] (Fig. 5). This led us to conclude that an early cellular function affected in-common by EOfAD mutations is energy production by mitochondria, a process that has been implicated in the pathological mechanisms underlying AD and many other neurodegenerative diseases previously (for reviews, see [41 –43]). Of course, identification of an expression gene set as significantly different between two genotypes does not easily identify the molecular process that is occurring. Metabolomics analysis of our most severe EOfAD-like mutation in psen1 (allele psen1^{Q96 _ K97del}, affecting the first luminal loop of the zebrafish Presenilin 1 protein where many relatively severe EOfAD mutations occur in the equivalent region of the human PSEN1 protein) has indicated decreased levels of fatty acids and, likely, decreased glutamine levels implying decreased ATP synthesis by mitochondria (McCubbin et al., manuscript in preparation). We are currently applying ¹³C-fluxomics approaches to identify the changes in central carbon metabolism in more detail.

Fig. 5

Reprint of Figure 1A,B from Barthelson et al. [9] . RNAseq analysis of 6-month-old zebrafish models of EOfAD. A) Schematic of an RNAseq experiment using zebrafish. A single mating of a single pair of fish heterozygous for either an EOfAD-like or a non-EOfAD-like mutation results in a family of heterozygous mutant, transheterozygous mutant and wild-type siblings. Comparisons made between genotypes in a RNAseq experiment are depicted. B) Heatmap summary of significantly altered KEGG [57] and IRE [50] gene sets in zebrafish EOfAD genetic models at 6 months of age. Only gene sets significantly altered (FDR-adjusted harmonic mean p < 0.05) in at least two comparisons of mutant zebrafish to their corresponding wild type siblings are shown. Columns are grouped by whether or not the zebrafish genotype is EOfAD-like, and rows are clustered based on their Euclidean distance. The numbers are FDR-adjusted harmonic mean p-values.

DISRUPTION OF IRON HOMEOSTASIS IN AD AND AD-LIKE MODELS IS SEEN THROUGH CHANGES IN EXPRESSION OF GENES THAT CONTAIN IRON RESPONSIVE ELEMENTS

Iron accumulates in AD brains with a possible mechanism for this being suggested by an analysis of a lysosomal storage disorder, Pompe disease. In 2019, Yambire et al. [44] presented evidence that lysosomal storage of glycogen due to loss of function of the gene GAA causes failure of correct lysosomal acidification. This led to accumulation of ferric iron (Fe³⁺) in lysosomes and a cytosolic deficiency of ferrous iron (Fe²⁺) with severe consequences for the biogenesis and function of mitochondria. Insufficient lysosomal acidification is also an effect-in-common of EOfAD mutations in APP [45] and PSEN1 [46], as well as due to the ɛ4 allele of APOE [47] and so would be expected to disturb cellular iron homeostasis.

The involvement of changes in iron homeostasis in EOfAD has long been implied by the discovery of an iron responsive element (IRE) in the 5’ untranslated region (UTR) of APP mRNA [48]. IREs are hairpin-loop structures in RNA that bind iron regulatory proteins, IRP1 and IRP2, both of which can sense deficiency of iron-sulphur clusters, often due to cellular deficiency of ferrous iron [48, 49]. The transcripts of over 10% of genes are thought to contain IREs, indicating the importance of iron homeostasis for cell function [50]. Our desire to investigate the interactions of EOfAD mutations with iron homeostasis led us to use the SIRES tool (“Searching for Iron Responsive ElementS” [51] –unfortunately the tool is no longer available online) to define lists of genes with putative IREs in their 5’ or 3’ UTRs for the zebrafish, mouse, and humans. At the time a common assumption in IRE research was that IREs in 5’ UTRs bound IRPs to inhibit translation, while IRPs bound to IREs in 3’ UTRs caused stabilization of mRNAs. Our comprehensive analysis of changes in the levels of IRE-containing mRNAs under various conditions overthrew the simplicity of this paradigm by showing that the presence of an IRE in either the 5’ or 3’ UTR of a mRNA could not, in itself, be used to predict the levels of the transcript. Whether an IRE in the 5’ or 3’ UTR of an mRNA gives increased or decreased transcript levels under ferrous iron replete or deficient conditions is an evolved function of that transcript and does not follow any simple positional rule.

Interestingly, our analysis of the behavior of IRE-containing transcripts in various systems in Hin et al. [50] revealed that postmortem AD brains show apparently increased levels of transcripts coding for the iron importation protein Transferrin Receptor (TFRC) consistent with AD brains experiencing a deficiency of ferrous iron. We also noted that,

“Remarkably, the “Blalock Alzheimer’s Disease Up” gene set from the C2 collection was significantly over-represented in both the sets of all 3' IREs (Bonferroni-adjusted Wilkinson’s p-value=3.9e–14) and all 5' IREs (Bonferroni adjusted Wilkinson’s p-value=2.1e–26) in the meta-analysis, and was also the gene set with the most significant over-representation of the human all 3' IREs set (Bonferroni-adjusted Fisher’s exact test p-value=7.8e–58) . . . This supports that disturbance of iron homeostasis particularly distinguishes AD from other disease conditions and pathways represented within the C2 collection.”

Our intra-family analysis of IRE-containing transcripts in 6-month-old zebrafish psen1^{Q96 _ K97del/+} compared to psen1^+/+ brains supported that EOfAD-like mutations in the presenilin genes cause disruption of iron homeostasis early in the development of AD [50].

THE FASCINATING PROBLEM OF CHROMOSOMALLY CO-LOCATED DIFFERENTIALLY EXPRESSED GENES

When transcriptome comparisons are conducted between sample groups that do not have identical genetic backgrounds (i.e., are not isogenic) then DE genes can appear due to differences in the rates of transcription and/or the stability of the transcripts of their alleles. Such genes influenced by “expression quantitative trait loci” (eQTLs) need not be DE due to the difference-of-interest between the samples (i.e., due to the presence of a mutation or a treatment being studied) [52]. These genes that are DE due to eQTLs therefore represent noise that can reduce the sensitivity of gene set enrichment analyses and lead to mistaken inferences regarding mutation or treatment effects. This problem is particularly common when using zebrafish for analysis of the transcriptomic effects of mutations as isogenic zebrafish strains are rarely used. When, for example, two heterozygous mutant fish are mated to produce WT, heterozygous, and homozygous mutant progeny, and then the transcriptomes of the WT and homozygous mutant fish are compared, the selection for the homozygous mutant genotype increases the likelihood of homozygosity for alleles of genes in linkage with the mutation and these alleles may differ in expression level from those on the homologous, WT chromosomes of the comparison sample group (see Fig. 6). The result is observation of a cluster of apparently DE genes around the mutated gene of interest –Chromosomally Co-located DE genes, CC-DEGs.

Fig. 6

Transcriptome analysis within non-isogenic zebrafish families reveals CC-DEGs. A) From Figure 1A in [53]. Inbreeding and selection for homozygous mutant and WT chromosomes creates homozygosity for eQTLs in cis with the mutation. B) From Figure 2A in [70]. Transcriptome comparison of 2 days post-fertilization embryos homozygous for a mutation in fmr1 with homozygous WT embryos sharing the same grandparents shows clustering of DE genes on Chromosome 14 that contains the fmr1 gene. Many of the DE genes are likely due to eQTLs.

When comparing the brain transcriptomes of our heterozygous EOfAD-like mutant fish to their WT siblings, we also noticed CC-DEGs and so made an effort to determine whether these were due to eQTLs or were DE due to the functional effects of the mutations. We developed a method for assessing genetic difference between genotype groups at the local level (i.e., at single nucleotide positions across the genome)—the differential allelic representation, DAR, metric—and used this to weight gene rankings in gene set enrichment analysis. This strategy increased the sensitivity of the analysis for detection of significant gene sets and allowed us to see that CC-DEGs may arise both due to eQTL effects but also due to apparent positive selection for advantageous linkage disequilibrium between functionally related genes over evolutionary timescales [53]. The CC-DEG effect is particularly marked in zebrafish since male zebrafish show lower rates of meiotic recombination compared to females so that most chromosomes in male zebrafish have genetic recombination lengths of less than 50 centiMorgans [54].

OTHER LESSONS LEARNED FROM TRANSCRIPTOME ANALYSIS IN ZEBRAFISH

Combining multiple gene set analysis approaches to achieve robust results

One of the most popular ways to perform gene set enrichment analysis is to test for over-representation of pre-defined gene sets such as gene ontology (GO) terms [55, 56] or KEGG pathways [57] within a set of differentially expressed genes. This method helps to identify biologically relevant consequences from changes in gene expression. However, due to the arbitrary hard threshold commonly applied to consider a gene as differentially expressed (i.e., an FDR-adjusted p-value of < 0.05), more subtle changes to gene expression, such as those caused by EOfAD-like mutations in zebrafish, may be difficult to identify.

Many alternative gene set enrichment methods do not rely on hard thresholds to consider a gene to be differentially expressed. GSEA (gene set enrichment analysis), originally conceptualized in 2005 [58], ranks all genes detectable in the experiment by differential expression. However, this method is highly susceptible to biases within the data [59]. Since then, other gene set testing methods have been developed. ROAST [60] and its fast implementation FRY [61], are focused gene set tests which use rotation (a Monte Carlo simulation technique) to determine whether gene sets are significantly altered. CAMERA [59] is a linear model-based gene set enrichment analysis technique which also estimates a correlation coefficient to account for inter-gene correlations. Currently, there is no gold standard for a single method which should be used for gene set enrichment analysis, as each method has its own strengths and weaknesses. Additionally, each method often results in different levels of statistical significance for a particular gene set [62].

We were inspired by the EGSEA [63] framework to combine the p-values of multiple methods of gene set enrichment analysis to give an overall statistical significance value for a particular gene set, whilst ensuring methods produce a diverse set of results and do not simply reinforce the results obtained by repeated testing of near-identical null hypotheses multiple times. For this reason, we chose to combine the raw p-values of FRY, CAMERA, and GSEA. The EGSEA framework default option of combining p-values is using Wilkinson’s method [64]. However, Wilkinson’s method may not be entirely appropriate, as it relies on the assumption that each p-value is independent. To overcome this limitation, our analyses utilized the harmonic mean p-value method, which has specifically been shown to be robust to dependent tests [65]. The intrinsic behavior of the harmonic mean results in a skewing of the resulting p-value towards the smallest value. However, it does not create more extreme p-values, unlike approaches such as Fisher’s method, and has the effect of moderating extreme p-values which already exist. To further protect from Type I errors, we apply other appropriate adjustments, such as Bonferroni or Benjamini-Hochberg’s FDR, to the resulting harmonic mean p-values.

In our experience, the use of the harmonic mean has allowed us to identify subtle, consistent, and biologically relevant changes to gene expression in EOfAD-like mutant zebrafish brains.

Homozygotes are not necessarily just more extreme heterozygotes

A common assumption made when using animal models of dominant EOfAD mutations is that homozygous (or transheterozygous/compound heterozygous) mutant animals will simply present a more severe version of a heterozygous phenotype. However, our intra-family analysis of two sorl1 mutations showed very different patterns of expression of genes of the KEGG_ALZHEIMERS_DISEASE gene set dependent on whether mutations were heterozygous or transheterozygous (i.e., no WT alleles present, Fig. 7). A subset of the genes in this pathway show consistently reduced expression in either heterozygous mutant, but are strongly increased in the transheterozygous mutants.

Fig. 7

Changes to gene expression in heterozygous and transheterozygous sorl1 fAD-like mutant zebrafish. Heatmap showing the logFC values for genes in the KEGG_ALZHEIMERS_DISEASE gene set in two sorl1 heterozygous mutant zebrafish brains (V1482Afs/+) and (R122Pfs/+), and in brains transheterozygous for these two alleles (i.e., sorl1 V1482Afs/R122Pfs). The logFC values were obtained from Additional File 5 in [37].

Mammalian ribosomal RNA depletion systems do not function well for zebrafish

Our early attempts at zebrafish brain transcriptome analysis were made unnecessarily problematic by use of mammalian sequence-based rRNA depletion kits. These failed to adequately deplete zebrafish rRNAs in our samples with rRNA molecules constituting highly variable rRNA proportions (10–35%) of the final sequenced libraries, which became easily confounded with experimental groups being analyzed, although this may also be a hallmark of the ribo-reduction kits of the era. We found evidence for differential abundance of k-mers (short sequences of 5–10bp) in these datasets, indicating the presence of sequence-specific biases as a consequence of variable depletion, with obvious implications for bias in differential expression analysis. An identical analysis on datasets where polyA enrichment was chosen over rRNA depletion showed no differential abundance of k-mers, supporting that differential expression of genes is not the cause of differential abundance of k-mers (unpublished). We subsequently moved to using polyA-based systems for library construction for RNAseq to minimize any confounding this highly variable step may introduce. While this means that information from some non-protein-coding transcripts is lost, this is generally compensated for by the lack of complications imposed due to variable rRNA concentrations. In the context of modelling a human disease state using a highly diverged species such as zebrafish, we also believe analysis of gene-level information from poly-adenylated transcripts is more likely to yield transferrable information, as the activities of many non-coding RNAs are less likely to be conserved across species (reviewed in [66]).

FUTURE WORK

Our program of creating EOfAD-like mutations in zebrafish and analyzing their young adult brains using transcriptomics to find the initial cellular stresses leading to AD identified changes in oxidative phosphorylation as common to all such mutations [9]. Interestingly, this has not been seen in transcriptome analyses of neurons differentiated from iPSCs bearing EOfAD mutations [67, 68]. This suggests that the oxidative phosphorylation changes may be an emergent property of brain structure, for example, due to vascular changes affecting brain oxygen perfusion). Alternatively, the oxidative phosphorylation signal in bulk brain RNAseq may be derived from a cell type other than neurons. Analysis of this will require a snRNAseq approach. One consistent feature between these iPSC analyses and our transcriptome analysis of 7 day-post-fertilization heterozygous psen1 EOfAD-like mutant zebrafish larvae [69] is a highly significant signal for changes in extracellular matrix component expression. This supports that the involvement of extracellular matrix in EOfAD should be a focus of future AD research.

CONCLUSION

In summary, our unique zebrafish models have illustrated the power of this model organism for understanding the earliest changes underlying AD, and their particular suitability in transcriptomic analyses. We have identified that changes to energy metabolism occur due to EOfAD-like mutations in these models at a young adult age. This may represent a cell function to be addressed with novel therapeutic strategies to prevent, or at least delay, the onset of AD.

Footnotes

ACKNOWLEDGMENTS

Michael Lardelli would like to acknowledge the repeated support of the Carthew Family Charity Trust over the years without which the Alzheimer’s Disease Genetics Laboratory could not have continued its pursuit of understanding of the genetic basis of familial Alzheimer’s disease.

FUNDING

ML is an academic employee of the University of Adelaide. SMP is a postdoctoral fellow supported by the Telethon Kids Institute. KB is supported by a Race Against Dementia –Dementia Australia Research Foundation Postdoctoral Fellowship and funds from Flinders University. LB and AA are supported by Australian Government Research Training Program Scholarships. NH is employed by Alkahest Inc. San Carlos, California.

CONFLICT OF INTEREST

Karissa Barthelson is an Editorial Board Member of this journal but was not involved in the peer-review process nor had access to any information regarding its peer-review.

DATA AVAILABILITY

Data sharing is not applicable to this article as no datasets were generated or analyzed during this reflective review. Please refer to the original publications cited for access to their data. Where unpublished work has been referred to, the data accompanying that work will be made available when a manuscript describing the work is placed on a preprint server.

References

Kawasumi

, Chiba

, Yamada

, Miyamae-Kaneko

, Matsuoka

, Nakahara

, Tomita

, Iwatsubo

, Kato

, Aiso

, Nishimoto

, Kouyama

(2004) Targeted introduction of V642I mutation in amyloid precursor protein gene causes functional abnormality resembling early stage of Alzheimer’s disease in aged mice. Eur J Neurosci 19, 2826–2838.

Sun

, Beglopoulos

, Mattson

, Shen

(2005) Hippocampal spatial memory impairments caused by the familial Alzheimer’s disease-linked presenilin 1 M146V mutation. Neurodegener Dis 2, 6–15.

Nakano

, Kondoh

, Kudo

, Imaizumi

, Kato

, Miyazaki

, Tohyama

, Takeda

(1999) Accumulation of murine amyloidbeta42 in a gene-dosage-dependent manner in PS1 ‘knock-in’ mice. Eur J Neurosci 11, 2577–2581.

McKean

, Handley

, Snell

(2021) A review of the current mammalian models of Alzheimer’s disease and challenges that need to be overcome. Int J Mol Sci 22, 13168.

Foley

, Ammar

, Lee

, Mitchell

(2015) Systematic review of the relationship between amyloid-beta levels and measures of transgenic mouse cognitive deficit in Alzheimer’s disease. J Alzheimers Dis 44, 787–795.

Hargis

, Blalock

(2017) Transcriptional signatures of brain aging and Alzheimer’s disease: What are our rodent models telling us? Behav Brain Res 322, 311–328.

Saito

, Matsuba

, Mihira

, Takano

, Nilsson

, Itohara

, Iwata

, Saido

(2014) Single App knock-in mouse models of Alzheimer’s disease. Nat Neurosci 17, 661–663.

Pang

, Jiang

, Zhang

, Yang

, Li

L-L

, Shimozawa

, Tambaro

, Mayer

, Zhang

, Li

, Wang

, Liu

, Yang

, Chen

, Liu

, Winblad

, Han

, Jiang

, Wang

, Nilsson

, Guo

, Lu

(2022) An App knock-in rat model for Alzheimer’s disease exhibiting Aβ and tau pathologies, neuronal death and cognitive impairments. Cell Res 32, 157–175.

Barthelson

, Newman

, Lardelli

(2022) Brain transcriptomes of zebrafish and mouse Alzheimer’s disease knock-in models imply early disrupted energy metabolism. Dis Model Mech 15, dmm049187.

10.

Park

, Cho

, Kim

, Saito

, Saido

, Won

, Kim

(2023) Single-cell RNA-sequencing identifies disease-associated oligodendrocytes in male APP NL-G-F and 5XFAD mice. Nat Commun 14, 802.

11.

Khalil

(2020) The genome editing revolution: review. J Genet Eng Biotechnol 18, 68.

12.

Doetschman

, Gregg

, Maeda

, Hooper

, Melton

, Thompson

, Smithies

(1987) Targetted correction of a mutant HPRT gene in mouse embryonic stem cells. Nature 330, 576–578.

13.

Thomas

, Capecchi

(1987) Site-directed mutagenesis by gene targeting in mouse embryo-derived stem cells. Cell 51, 503–512.

14.

Oblak

, Forner

, Territo

, Sasner

, Carter

, Howell

, Sukoff-Rizzo

, Logsdon

, Mangravite

, Mortazavi

, Baglietto-Vargas

, Green

, MacGregor

, Wood

, Tenner

, LaFerla

, Lamb

, and The MODEL-AD; Consortium (2020) Model organism development and evaluation for late-onset Alzheimer’s disease: MODEL-AD. Alzheimers Dement (N Y) 6, e12110.

15.

Gemberling

, Bailey

, Hyde

, Poss

(2013) The zebrafish as a model for complex tissue regeneration. Trends Genet 29, 611–620.

16.

Hin

, Newman

, Kaslin

, Douek

, Lumsden

, Nik

SHM

, Dong

, Zhou

, Manucat-Tan

, Ludington

, Adelson

, Pederson

, Lardelli

(2020) Accelerated brain aging towards transcriptional inversion in a zebrafish model of the K115fs mutation of human PSEN2. PLoS One 15, e0227258.

17.

Chia

, Klingseisen

, Sieger

, Priller

(2022) Zebrafish as a model organism for neurodegenerative disease. Front Mol Neurosci 15, 940484.

18.

Musa

, Lehrach

, Russo

(2001) Distinct expression patterns of two zebrafish homologues of the human APP gene during embryonic development. Dev Genes Evol 211, 563–567.

19.

Otis

, Zeituni

, Thierer

, Anderson

, Brown

, Boehm

, Cerchione

, Ceasrine

, Avraham-Davidi

, Tempelhof

, Yaniv

, Farber

(2015) Zebrafish as a model for apolipoprotein biology: comprehensive expression analysis and a role for ApoA-IV in regulating food intake. Dis Model Mech 8, 295–309.

20.

Moore

, Gillentine

, Botezatu

, Wilson

, Benson

, Langeland

(2014) Asynchronous evolutionary origins of Abeta and BACE1. Mol Biol Evol 31, 696–702.

21.

Bhattarai

, Thomas

, Cosacak

, Papadimitriou

, Mashkaryan

, Froc

, Reinhardt

, Kurth

, Dahl

, Zhang

, Kizil

(2016) IL4/STAT6 signaling activates neural stem cell proliferation and neurogenesis upon amyloid-beta42 aggregation in adult zebrafish brain. Cell Rep 17, 941–948.

22.

Bhattarai

, Thomas

, Cosacak

, Papadimitriou

, Mashkaryan

, Zhang

, Kizil

(2017) Modeling amyloid-beta42 toxicity and neurodegeneration in adult zebrafish brain. J Vis Exp, 56014.

23.

Cosacak

, Bhattarai

, Reinhardt

, Petzold

, Dahl

, Zhang

, Kizil

(2019) Single-cell transcriptomics analyses of neural stem cell heterogeneity and contextual plasticity in a zebrafish brain model of amyloid toxicitye. Cell Rep 27, 1307–1318e1303.

24.

Sharman

, Moussavi Nik

, Chen

, Ong

, Wijaya

, Laws

, Taddei

, Newman

, Lardelli

, Martins

, Verdile

(2013) The guinea pig as a model for sporadic Alzheimer’s disease (AD): the impact of cholesterol intake on expression of AD-related genes. PLoS One 8, e66235.

25.

Moussavi Nik

, Newman

, Wilson

, Ebrahimie

, Wells

, Musgrave

, Verdile

, Martins

, Lardelli

(2015) Alzheimer’s disease-related peptide PS2V plays ancient, conserved roles in suppression of the unfolded protein response under hypoxia and stimulation of gamma-secretase activity. Hum Mol Genet 24, 3662–3678.

26.

Sato

, Imaizumi

, Manabe

, Taniguchi

, Hitomi

, Katayama

, Yoneda

, Morihara

, Yasuda

, Takagi

, Kudo

, Tsuda

, Itoyama

, Makifuchi

, Fraser

, St George-Hyslop

, Tohyama

(2001) Increased production of beta-amyloid and vulnerability to endoplasmic reticulum stress by an aberrant spliced form of presenilin 2. J Biol Chem 276, 2108–2114.

27.

Katayama

, Imaizumi

, Manabe

, Hitomi

, Kudo

, Tohyama

(2004) Induction of neuronal death by ER stress in Alzheimer’s disease. J Chem Neuroanat 28, 67–78.

28.

Kumar

, Hedges

(1998) A molecular timescale for vertebrate evolution. Nature 392, 917–920.

29.

Crow

, Suresh

, Lee

, Gillis

(2022) Coexpression reveals conserved gene programs that co-vary with cell type across kingdoms. Nucleic Acids Res 50, 4302–4314.

30.

Wiley

, Redfield

, Zon

(2017) Chemical screening in zebrafish for novel biological and therapeutic discovery. Methods Cell Biol 138, 651–679.

31.

Adatto

, Lawrence

, Krug

, Zon

(2022) The effects of intensive feeding on reproductive performance in laboratory zebrafish (Danio rerio). PLoS One 17, e0278302.

32.

Barthelson

, Pederson

, Newman

, Lardelli

(2021) Brain transcriptome analysis of a protein-truncating mutation in sortilin-related receptor 1 associated with early-onset familial Alzheimer’s disease indicates early effects on mitochondrial and ribosome function. J Alzheimers Dis 79, 1105–1119.

33.

Barthelson

, Dong

, Newman

, Lardelli

(2021) PRESENILIN 1 mutations causing early-onset familial Alzheimer’s disease or familial acne inversa differ in their effects on genes facilitating energy metabolism and signal transduction. J Alzheimers Dis 82, 327–347.

34.

Wang

, Yang

, Wen

, Sun

, Su

, Liu

, Ma

, Lv

, Wen

, Qu

, Chen

, Sun

, Shen

, Zhang

(2010) Gamma-secretase gene mutations in familial acne inversa. Science 330, 1065.

35.

Ishikawa

, Piao

, Miyashita

, Kuwano

, Onodera

, Ohtake

, Suzuki

, Nishizawa

, Takahashi

(2005) A mutant PSEN1 causes dementia with Lewy bodies and variant Alzheimer’s disease. Ann Neurol 57, 429–434.

36.

Barthelson

, Pederson

, Newman

, Jiang

, Lardelli

(2021) In-frame and frameshift mutations in zebrafish presenilin 2 affect different cellular functions in young adult brains. J Alzheimers Dis Rep 5, 395–404.

37.

Barthelson

, Pederson

, Newman

, Lardelli

(2020) Brain transcriptome analysis reveals subtle effects on mitochondrial function and iron homeostasis of mutations in the SORL1 gene implicated in early onset familial Alzheimer’s disease. Mol Brain 13, 142.

38.

Newman

, Hin

, Pederson

, Lardelli

(2019) Brain transcriptome analysis of a familial Alzheimer’s disease-like mutation in the zebrafish presenilin 1 gene implies effects on energy production. Mol Brain 12, 43.

39.

Jiang

, Newman

, Lardelli

(2018) The zebrafish orthologue of familial Alzheimer’s disease gene PRESENILIN 2 is required for normal adult melanotic skin pigmentation. PLoS One 13, e0206155.

40.

Jiang

, Pederson

, Newman

, Dong

, Barthelson

, Lardelli

(2020) Transcriptome analysis indicates dominant effects on ribosome and mitochondrial function of a premature termination codon mutation in the zebrafish gene psen2. PLoS One 15, e0232559.

41.

Noori

, Mezlini

, Hyman

, Serrano-Pozo

, Das

(2021) Systematic review and meta-analysis of human transcriptomics reveals neuroinflammation, deficient energy metabolism, and proteostasis failure across neurodegeneration. Neurobiol Dis 149, 105225.

42.

, Jin

, Xu

, Zhu

(2022) Aberrant energy metabolism in Alzheimer’s disease. J Transl Int Med 10, 197–206.

43.

Ozgen

, Krigman

, Zhang

, Sun

(2022) Significance of mitochondrial activity in neurogenesis and neurodegenerative diseases. Neural Regen Res 17, 741–747.

44.

Yambire

, Rostosky

, Watanabe

, Pacheu-Grau

, Torres-Odio

, Sanchez-Guerrero

, Senderovich

, Meyron-Holtz

, Milosevic

, Frahm

, West

, Raimundo

(2019) Impaired lysosomal acidification triggers iron deficiency and inflammation in vivo. Elife 8, e51031.

45.

Jiang

, Sato

, Im

, Berg

, Bordi

, Darji

, Kumar

, Mohan

, Bandyopadhyay

, Diaz

, Cuervo

, Nixon

(2019) Lysosomal dysfunction in Down syndrome is APP-dependent and mediated by APP-betaCTF (C99). J Neurosci 39, 5255–5268.

46.

Lee

, Yu

, Kumar

, Lee

, Mohan

, Peterhoff

, Wolfe

, Martinez-Vicente

, Massey

, Sovak

, Uchiyama

, Westaway

, Cuervo

, Nixon

(2010) Lysosomal proteolysis and autophagy require presenilin 1 and are disrupted by Alzheimer-related PS1 mutations. Cell 141, 1146–1158.

47.

Prasad

, Rao

(2018) Amyloid clearance defect in ApoE4 astrocytes is reversed by epigenetic correction of endosomal pH. Proc Natl Acad Sci U S A 115, E6640–E6649.

48.

Kennedy

, Mende-Mueller

, Blondin

, Beinert

(1992) Purification and characterization of cytosolic aconitase from beef liver and its relationship to the iron-responsive element binding protein. Proc Natl Acad Sci U S A 89, 11730–11734.

49.

Terzi

, Sviderskiy

, Alvarez

, Whiten

, Possemato

(2021) Iron-sulfur cluster deficiency can be sensed by IRP2 and regulates iron homeostasis and sensitivity to ferroptosis independent of IRP1 and FBXL5. Sci Adv 7, eabg4302.

50.

Hin

, Newman

, Pederson

, Lardelli

(2021) Iron responsive element-mediated responses to iron dyshomeostasis in Alzheimer’s disease. J Alzheimers Dis 84, 1597–1630.

51.

Campillos

, Cases

, Hentze

, Sanchez

(2010) SIREs: searching for iron-responsive elements. Nucleic Acids Res 38, W360–367.

52.

White

, Mackay

, Wilson

, Busch-Nentwich

(2022) Allele-specific gene expression can underlie altered transcript abundance in zebrafish mutants. Elife 11, e72825.

53.

Baer

, Barthelson

, Postlethwait

, Adelson

, Pederson

, Lardelli

Differential allelic representation (DAR) identifies candidate eQTLs and improves transcrip- tome analysis. bioRxiv, 2023.2003.2002.530865.

54.

Singer

, Perlman

, Yan

, Walker

, Corley-Smith

, Brandhorst

, Postlethwait

(2002) Sex-specific recombination rates in zebrafish (Danio rerio). Genetics 160, 649–657.

55.

Gene Ontology

(2021) The Gene Ontology resource: enriching a GOld mine. Nucleic Acids Res 49, 334–D325-D.

56.

Ashburner

, Ball

, Blake

, Botstein

, Butler

, Cherry

, Davis

, Dolinski

, Dwight

, Eppig

, Harris

, Hill

, Issel-Tarver

, Kasarskis

, Lewis

, Matese

, Richardson

, Ringwald

, Rubin

, Sherlock

(2000) Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat Genet 25, 25–29.

57.

Kanehisa

, Goto

(2000) KEGG: Kyoto Encyclopedia of Genes and Genomes. Nucleic Acids Res 28, 27–30.

58.

Subramanian

, Tamayo

, Mootha

, Mukherjee

, Ebert

, Gillette

, Paulovich

, Pomeroy

, Golub

, Lander

, Mesirov

(2005) Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 102, 15545–15550.

59.

, Smyth

(2012) Camera: a competitive gene set test accounting for inter-gene correlation. Nucleic Acids Res 40, e133.

60.

, Lim

, Vaillant

, Asselin-Labat

, Visvader

, Smyth

(2010) ROAST: rotation gene set tests for complex microarray experiments. Bioinformatics 26, 2176–2182.

61.

Ritchie

, Phipson

, Wu

, Hu

, Law

, Shi

, Smyth

(2015) limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res 43, e47–e47.

62.

Geistlinger

, Csaba

, Santarelli

, Ramos

, Schiffer

, Turaga

, Law

, Davis

, Carey

, Morgan

, Zimmer

, Waldron

(2021) Toward a gold standard for benchmarking gene set enrichment analysis. Brief Bioinform 22, 545–556.

63.

Alhamdoosh

, Ng

, Wilson

, Sheridan

, Huynh

, Wilson

, Ritchie

(2017) Combining multiple tools outperforms individual methods in gene set enrichment analyses. Bioinformatics 33, 414–424.

64.

Wilkinson

(1951) A statistical consideration in psychological research. Psychol Bull 48, 156–158.

65.

Wilson

(2019) The harmonic meanp-value for combining dependent tests. Proc Natl Acad Sci U S A 116, 1195–1200.

66.

Johnsson

, Lipovich

, Grander

, Morris

(2014) Evolutionary conservation of long non-coding RNAs; sequence, structure, function. Biochim Biophys Acta 1840, 1063–1071.

67.

Kwart

, Gregg

, Scheckel

, Murphy

, Paquet

, Duffield

, Fak

, Olsen

, Darnell

, Tessier-Lavigne

(2019) A large panel of isogenic APP and PSEN1 mutant human iPSC neurons reveals shared endosomal abnormalities mediated by APP beta-CTFs, not Abeta. Neuron 104, 1022.

68.

Corsi

, Gadekar

, Haukedal

, Doncheva

, Anthon

, Ambardar

, Palakodeti

, Hyttel

, Freude

, Seemann

, Gorodkin

(2023) The transcriptomic landscape of neurons carrying PSEN1 mutations reveals changes in extracellular matrix components and non-coding gene expression. Neurobiol Dis 178, 105980.

69.

Dong

, Newman

, Pederson

, Barthelson

, Hin

, Lardelli

(2021) Transcriptome analyses of 7-day-old zebrafish larvae possessing a familial Alzheimer’s disease-like mutation in psen1 indicate effects on oxidative phosphorylation, ECM and MCM functions, and iron homeostasis. BMC Genomics 22, 211.

70.

Barthelson

, Baer

, Dong

, Hand

, Pujic

, Newman

, Goodhill

, Richards

, Pederson

, Lardelli

(2021) Zebrafish chromosome 14 gene differential expression in the fmr1hu2787 model of fragile X syndrome. Front Genet 12, 723.