Abstract
Embryonic growth trajectory is a risk factor for chronic metabolic and cardiovascular disorders and influences birth weight along with early post-natal weight gain in humans. Grb10 is a negative regulator of the main pathways driving embryonic growth and knock-out in mammals increases insulin sensitivity and growth trajectory. This study investigates in Danio rerio the long-term cardiometabolic consequences and associated transcriptomic profiles of morpholino induced early life disruption of grb10a expression. The associated transient knockdown of grb10a increased embryonic growth (+7 %) and metabolic rate (+25 %), while decreasing heart rate (−50 %) in early life. Juvenile growth and respiratory rate were also elevated (+30 % and 7-fold increase respectively). This was associated with permanent remodelling of the transcriptional landscape and the dysregulation of multiple growth-related pathways. This study indicates that zebrafish are a suitable model for life-long investigation of the link between early growth and later life disease risk.
Introduction
In humans, small and large for gestational age (SGA, LGA) status, caused by embryonic growth aberration, are early risk factors for chronic metabolic and cardiovascular disorders, such as type II diabetes (TIID), obesity, cardiac dysfunction, and hypertension (1–3). As these disorders are an ever-increasing concern, efforts must be made to identify early risk factors and develop novel therapeutics for prevention and treatment. While this correlation is widely replicated in human populations, targeted in-vivo studies to elucidate this phenomenon and the pathways involved are lacking. The zebrafish is an ideal model organism for longitudinal study owing to its rapid generation time and is also highly suitable for developmental studies as embryos are easily accessible and transparent.
Embryonic growth is driven by the insulin/insulin like growth factor (Ins/IGF) signaling pathway (4). Circulating insulin and IGF levels indicate the caloric and nutrient condition of the extrauterine environment and developmental rate positively correlate with Ins/IGF serum levels. Glucose metabolism, glycogen and protein synthesis, as well as proliferation are promoted if the environment is considered favourable (5). Developmental plasticity enables phenotypic alterations to occur in response to fluctuations in the external environment to ensure survival. This is important for healthy development, though phenotypic responses for immediate survival may come at the expense of elevated disease risk in later life. While dysfunction of the Ins/IGF pathway correlates with growth restriction and dysfunctional energy homeostasis (6), SGA and LGA individuals rarely present with mutations in the core pathway, suggesting other factors must be involved in subsequent pathway deregulation.
Growth factor receptor bound protein 10 (GRB10) is a negative regulator of the Insulin/IGF signaling pathway. Inhibition by GRB10 downregulates the growth response, promoting a switch from glucose to fat metabolism and halting cell cycle progression (7–9). GRB10 expression limits placental growth and efficiency (10) and correlates with small body size (11). In humans, genome wide association studies show GRB10 is strongly associated with TIID (12), and GRB10 copy number variation is associated with Silver Russell Syndrome (13), a rare growth disorder typically characterised by intrauterine growth restriction, SGA status, hypoglycemia, poor muscle development, and fat deposition. The role of GRB10 in the regulation of human growth is notably associated with response to recombinant human growth hormone in children with growth hormone deficiency (14,15). The interaction between the environment and the control of GRB10 expression has also been investigated, and growth response was shown to be modulated by genotype in association with day length (16). As metabolic and cardiac disorders presenting in adulthood are thought to originate during foetal development (17–19), understanding the impact of altered embryonic growth trajectory on later life disease is important to highlight at-risk individuals.
Moreover, variability in Grb10 expression is also linked to the dramatic range of body sizes observed between cetacean species (20), suggesting that modulating grb10 expression could be an important target for agri- and aqua-culture. Average daily mass gain is elevated in beef cattle with a grb10 associated deletion (21), and grb10 single nucleotide polymorphisms (SNPs) impact muscle and lipid mass, angularity, and body conditioning score (22). Global knockout of grb10 in mice correlates with elevated muscle mass, reduced lipid mass, and insulin sensitivity (23,24). Therefore, understanding the impact disruptions to early-life growth trajectory have on mature organism size, average daily gain, and lipid to muscle ratio may provide the groundwork to boosting meat yield, and quality which is required to match the growing global demand for protein.
Here, zebrafish (Danio rerio) were used to assess the importance of grb10 as a coordinator of growth, metabolism, and cardiac health, in order to investigate the impact of early-life growth disruption on later-life cardiometabolic phenotype. Grb10a was transiently suppressed during zebrafish embryogenesis by antisense oligonucleotide directed blocking of mRNA splicing. Grb10a was confirmed to have an impact not only on embryonic growth, but also on cardiac and metabolic phenotypes. Importantly, this impact was found to persist into adulthood, inducing an enduring remodelling of the transcriptomic landscape.
Materials and Methods
Zebrafish Husbandry
AB zebrafish were maintained under standard conditions (≈28 °C; 14 h light/10 h dark cycle; stocking density < 5 fish per litre) within the Biological Services Unit of The University of Manchester. Embryos were collected and raised in egg water (Instant Ocean salt 60 µg/mL) up to 5 days post-fertilisation (dpf) and transferred to the main aquarium system. Young fry were fed powdered food and rotifers, while older fry and adults were fed powdered food and brine shrimp (ZM Fish Food and Equipment, Winchester, UK). All regulated procedures received ethical approval from the Institution’s ethical review board and were performed under a Home Office Licence (PPL P005EFE9F9 to HAS).
Transient Knockdown of grb10a Expression by Microinjection of Antisense Oligonucleotide
In accordance with current guidelines for morpholino use in zebrafish (25), the use of morpholino-modified antisense oligonucleotides to knockdown (KD) grb10a was validated in several ways. Morpholinos targeting two distinct sites should yield a similar phenotype, which should display a dose-response effect and be rescuable by RNA co-injection, and a suitable control morpholino with no phenotypic effect should be used to account for microinjection technique (25). Morpholino-modified antisense oligonucleotides targeting the splice donor sites of zebrafish grb10a coding exon three (e3i3) or four (e4i4) were designed by and obtained from Gene Tools, LLC (Philomath, OR, USA) along with a standard control (SC) oligonucleotide targeting human β-globin. Oligonucleotide sequences are listed in Table 1. Microinjection solutions were generated as outlined in Table 2. Concentrations of ingredients were consistent with standard practice (25). Phenol red and n-Cerulean (nuclear-targeting blue fluorescent protein) mRNA were included in each solution to allow monitoring of successful injection. A P-97 Flaming/Brown Micropipette Puller (Sutter Instruments, Novato, CA, USA) was used to generate microinjection needles from 1 mm OD glass capillaries, which were backloaded with 2 μl of injection solution and mounted on a needle holder held in a micromanipulator attached to a PLI-100A Picoliter Injector (Warner Instruments, Hamden, CT, USA). Embryos were injected once, as per established method (25), into the yolk directly below the cell mass. Embryos were injected before the four-cell stage and were screened for fluorescence at 48 hours post fertilisation (hpf) to ensure constitutive and even uptake of the injection material. Non-uniformly or weakly-fluorescent embryos were removed.
Oligonucleotide sequences
Microinjection Solutions
Validation of grb10a Knockdown by PCR
Primer sequences, outlined in Table 1, were designed using SnapGene® software (GSL Biotech, San Diego, CA, USA; available at snapgene.com) and synthesised by Thermo Fisher Scientific (Waltham, MA, USA). Sequences were designed to be approximately 20 base pairs with a GC content between 40 and 50 % and start with GC or GG to improve target annealing. Primer pairs were designed to anneal at similar temperatures, and specificity was confirmed with Primer BLAST (26). To confirm antisense oligonucleotide activity, RNA was extracted from pooled zebrafish embryos at 24, 48, 72, 96, and 120 hpf (n=3, 5 embryos per pool). RNA extraction was performed using QIAGEN RNeasy lipid extraction kit according to the manufacturer’s instructions, and cDNA was generated by reverse transcription using the ProtoScript® II First Strand cDNA Synthesis Kit (NEB). The cDNA was amplified with primers flanking each splice site. PCR was performed using Taq polymerase (New England Biolabs (NEB, Hitchin, UK)) with thermocycling parameters outlined in Table 3. β-actin (actb1) was used as a control for cDNA integrity.
Thermocycling Parameters
Signalling Analysis by Western Blotting
96 hpf zebrafish embryos were deyolked in Ringer’s Buffer (n=15, performed in triplicate). Embryos were resuspended in 100 μl RIPA buffer (150 mM NaCl, 1 % Nonident P-40, 0.5 % Sodium deoxycholate, 0.1 % SDS, 25 mM Tris pH 7.4) containing protease and phosphatase inhibitors (Thermo Fisher Scientific) and homogenised by passing through a 22 g needle. Samples were incubated on ice for 30 minutes. The supernatant was clarified by centrifugation at 13000 rpm at 4 °C for 20 minutes and collected. Samples were denatured at 98 °C for 5 minutes with an equal volume of Laemmli buffer (final concentration 2 % SDS, 10 % glycerol, 60 mM Tris-Cl, 0.01 % bromophenol blue, 0.1 % β-Mercaptoethanol). Proteins were separated on a 10 % SDS acrylamide gel using BioRad Western Blot apparatus and powerpack for one hour at 130 V, 400 mA (running buffer 25 mM Tris, 200 mM glycine, 0.03 % SDS). Proteins were transferred for one hour in transfer buffer (25 mM Tris-HCl pH 7.6, 200 mM glycine, 20 % methanol, 0.03 % SDS) onto a membrane preactivated in methanol. The membrane was washed three times in TBS-T (137 mM NaCl, 2.7 mM KCl, 19 mM Tris, 0.1 % Tween-20) prior to incubation in blocking buffer (3 % BSA in TBS-T) for one hour. The membrane was rinsed three times in TBS-T and incubated overnight in primary antibody (Table 4) at 4 °C under constant agitation. The membrane was rinsed three times in TBS-T prior to incubation in secondary antibody (Table 4) for one hour at room temperature. The membrane was then covered with ECL Western Blotting Substrate (Promega, Southampton, UK) and imaged immediately using a BioRad ChemiDoc XRS+ Imager. Protein expression was quantified from band intensity in ImageJ (27).
Western Blot Antibodies
Generation of grb10a mRNA for Overexpression and Knockdown Rescue
Total RNA was extracted from a pool of 96 hpf zebrafish embryos using QIAGEN RNeasy lipid extraction kit according to the manufacturer’s instructions, and cDNA was generated by reverse transcription using the ProtoScript® II First Strand cDNA Synthesis Kit (NEB). Grb10a was amplified from cDNA by high specificity PCR (NEB Q5 Hot Start) with primers complimentary to the 5’ and 3’ ends of the open reading frame (Table 1). The product was purified using QIAGEN Quick Gel Extraction, blunt ligated into pCR-Blunt II-TOPO (Thermo Fisher Scientific) and transformed into E. coli following an established protocol. Purified plasmid was sequenced by GATC Biotech Sanger Sequencing to confirm successful cloning. The insert was then liberated by digestion with Cla I and Xba I restriction enzymes (NEB) and subcloned in pCS2+(28). Capped RNA was generated using the mMESSAGE mMACHINE® SP6 Transcription Kit (Thermo Fisher Scientific) according to the manufacturer’s instructions. pCS2+ -grb10a was linearised by Nsi I (NEB) digestion prior to RNA generation. RNA was purified by MEGAclear™ Transcription Clean-Up. 1 μl of the purified RNA was analysed by gel electrophoresis to confirm amplification and structural integrity.
Physiological and Metabolic Phenotype Measurements
Whole body length, the longest straight-line distance between the snout and tip of the notochord, and yolk area measurements were taken every 24 hours from 48 to 120 hpf (Zeiss Axiocam MRm s1381, Zeiss ZEN lite (2012)). Embryos were dechorionated and acclimatised for ≥ one hour before imaging. Images were imported into ImageJ, which was calibrated with an image of a graticule of known size. Data were imported into GraphPad Prism version 7.00 for Windows (GraphPad Software, La Jolla, CA, USA, www.graphpad.com).
To investigate embryonic cardiac phenotype, embryos were sedated using a concentration known to have no impact on cardiac function (0.04 % MS-222 solution (29)). Heart beats were counted over a 20 second period and converted to beats per minute (bpm).
A Glucose Uptake-Glo TM Assay (Promega) was performed on 96 hpf zebrafish as an index of metabolic rate differences between treatment groups. Individual zebrafish embryos (n=5 per treatment) were injected with 1 mM 2-deoxyglucose-6-phosphate directly into the yolk and allowed to recover for 30 minutes. Embryos were processed in accordance with the manufacturer’s instructions. Luminescence was measured on a Synergy™ H1 Microplate reader (BioTek Instruments, Inc., Winooski, VT, USA), software version 2.07.17, with 8 readings per well.
Quantitative PCR
Quantitative PCR (qPCR) primers amplifying grb10a and markers of cardiac dysfunction were designed (Table 1) and their efficiency validated (30). RNA for comparison of gene expression between SC and KD samples was extracted from pooled (n=10) embryonic zebrafish or pooled (n=3) adult (<1 year) heart samples by QIAGEN RNEasy Lipid Tissue Extraction kit according to the manufacturer’s instructions. Samples were repeated in triplicate and tested for gene expression by qPCR (Applied Biosystems Power SyBr Green) (Table 3). Results were visualised using MxPro QPCR Software (Agilent, Santa Clara, CA, USA). Relative fold change in gene expression was calculated using the ΔΔCt method according to the following equation:
Where
And
Data were imported into GraphPad and unpaired t-tests were used to assess the difference between samples.
Transcriptomic Analysis
Temporal changes in gene expression were assessed by computational analysis of transcriptomic data. Pooled samples (n=5) were generated from each treatment group (SC, KD) at 5, 10, 15, 20, and 30 dpf, with three repeats per sample. Individuals were culled under terminal anaesthesia with only tissue anterior to the gills included, and total RNA was prepared. Transcriptomic data were generated using Affymetrix Zebgene 1.0st arrays. Data for all 75212 gene probes were imported into Qlucore Omics Explorer 2.2 (Lund, Sweden) as .cel files and normalised using the robust multi-array average (RMA) approach with a gene level summary. Zebrafish gene identities were assigned using the Affymetrix gene definitions. Human orthologues were mapped (GRCh 38) using the biomaRt R-package (31). A workflow pipeline of the subsequent transcriptomic analyses is outlined in Supplementary Figure 1.
Unsupervised analysis of gene expression by age group was conducted by generating hierarchically clustered heat maps. Standard deviation filtering (standard deviation of specific gene expression divided by maximum gene standard deviation [s/smax]) was performed on the dataset to remove genes with low variance, unlikely to be informative (or below detectable levels). The projection score (32), was used to determine the threshold for filtering by calculating the maximum separation present in principal component analysis (PCA).
Hypernetwork Modelling
Hypernetwork analysis was performed to model the dynamics of the transcriptome by quantifying shared correlations between target genes, a general model of which is outlined in Supplementary Figure 1. All analyses were performed in R (version 3.4.2).
Using samples from all time-points in either SC (n=15) or KD (n=15) fish, Pearson’s correlation coefficients (r) were calculated between genes identified by unsupervised analysis (g) [KD = 119, SC = 297] and the rest of the transcriptome (gc) [KD = 75093, SC= 74915]. Positive and negative correlations greater than ±1 standard deviation (sd) from the mean of the R-values were binarized separately and the resulting matrices combined. In this way, a matrix (M) was formed describing the positive and negative correlates of each element (∈) of g separately:
The resulting binary matrix (M) was multiplied by its transpose (Mt) to generate a matrix (M. Mt) which lists the number of shared correlations between any pair of genes (g). This matrix represents the hypernetwork describing the higher order relationships between genes not represented in traditional transcriptomic approaches. This measurement of co-ordination has been hypothesised to model functional relationships (33). Coordination between genes with age-related expression and the rest of transcriptome was determined using the Galois correspondence. This was represented by the subset of the transcriptome (⊂ gc) showing correlated expression with all the transcripts from the hypernetwork cluster (⊂ g).
Quantification of Network Topology
Correlation networks model functional relationships within gene networks (34) and allow identification of clusters of highly connected genes (35). Quantification of SC and KD hypernetwork properties (connectivity and entropy) was performed on the matrix M. Mt. Connectivity is defined here as the sum of connections shared by each element of the network (g) and measures the direct and indirect relationships between them.
Entropy describes the degree of disorder within the distribution of shared correlations measured in each element of g. Entropy is correlated with the differentiation potential of a cell (36), with low entropy indicating specific pathway activation and high entropy indicating high signalling pathway redundancy. Entropy has also been used as an index of regularity and patterning (37). Entropy was measured using R package BioQC (38).
As the hypernetworks formed three clusters of genes when clustered hierarchically, all following analysis focused on the 20-30 dpf cluster (86 SC genes, 67 KD genes). Clusters were isolated and within-cluster connectivity and entropy were calculated per gene.
The expected background connectivity and entropy were measured in 1000 iterated hypernetworks using randomly selected genes (86 random genes and 67 random genes for the SC and KD respectively) (Supplementary Figure 2). The median expected connectivity and entropy were subtracted from the experimentally derived values to control for the background. Experimental data for SC and KD zebrafish are therefore presented as the difference between expected and observed connectivity and entropy.
Gene Ontology
To associate gene expression with biological processes, gene set enrichment analysis (GSEA) was performed (39). Unsupervised analysis was supported with GSEA ranked by age group ANOVA false discovery rate modified p-values [q-values] (Qlucore) using genes mapped to human orthologs. GSEA was also performed on regression data (ranked by R-value) using Webgestalt (40). Over-representation analysis (ORA) was used to identify gene ontology associated with unranked gene sets (Webgestalt). All gene ontology analysis used the GO Biological Process Ontology gene list (Molecular Signatures Database (39,41)).
Oxygen Consumption as a Measure of Metabolic Rate
To analyse metabolic rate, oxygen consumption was measured in 30 dpf zebrafish from each treatment group. Individual zebrafish were placed into 1 of 4 stop-flow respirometry chambers (volume 2 ml) and allowed to acclimate at 28 °C for > one hour. Chambers were refreshed immediately prior to the experiment until oxygen saturation measured 100 %. Optical oxygen sensors paired with oxygen sensor spots (Pyroscience, Aachen, Germany) were used to measure oxygen saturation within the chambers and recorded using a FireStingO2 Fiber-optic oxygen and temperature meter. Temperature was maintained at 28 ± 0.3 °C and was monitored and recorded with the FireSting integrated temperature meter. Probes were calibrated according to the manufacturer’s instructions. Chambers were randomised per trial, with one chamber left empty to correct for background bacterial respiration. Oxygen consumption over time curves were recorded in triplicate with five trials per treatment. Chambers were refreshed when oxygen saturation reached 80 %. Oxygen consumption was measured as the change in oxygen saturation within the chamber per trial. To calculate the rate of oxygen consumption, linear regression in Microsoft Excel was used to approximate the change in oxygen saturation during each trial. This was normalised against the dry mass of the subject, length of the trial, and volume of the respirometry chamber according to the following equation:
Where 7900 μgL-1 is equivalent to 100 % oxygen saturation at 28 ° C. This normalises the data against the dry mass of the subject, time of the trial, and volume of the respirometry chamber.
Statistical Tests
For transcriptomic analyses, rank regression (least squares method) was used to generate the most appropriate linear model for each probe (the smallest degree of variance over the sample). Multi group analysis of variance (ANOVA) was used to associate each gene ID with time dependent gene expression. Wilcoxon rank sum test (ggpubR package for R (42)) was used to test for differences of network topology between groups. False discovery rate (FDR) adjustment was made using the Benjamini-Hochberg method and applied to the gene expression and ontology analysis (43).
All data were ROUT tested (44) for outliers and subject to D’Agostino and Pearson normality tests. All comparisons between SC and KD measurements were performed using an unpaired t-test. Comparisons of multiple groups were performed using one-way ANOVAs. Post-hoc power calculations were performed to confirm sample sizes were sufficient, where α = 0.05.
Results
Knockdown of grb10a Expression by Splice-Blocking Antisense Oligonucleotide
As grb10 has been linked to embryonic growth trajectory, expression was examined over the first 120 hpf (hours post fertilisation) in zebrafish embryos. qPCR analysis of grb10a expression at 24-hour intervals revealed a strong upregulation at 48 hpf compared to housekeeping genes (Figure 1a). Expression of the grb10 paralogue, grb10b, was not detectable at any time point.
1a. Grb10a qPCR of WT embryos (24-120 hpf, triplicated, n=5 per well). Data is shown as gene expression relative to β-actin and shows a significant peak at 48 hpf. 1b. Schematic of the first five exons of the zebrafish grb10a gene. 5’ splice sites are highlighted with the forward and reverse primer triad indicated. 1c. PCR amplification of the e3i3 and e4i4 splice site in embryos treated with either Standard Control morpholino, e3i3, or e4i4 (24-120hpf, n=5 per well). β-actin was used as a positive control. 1d. Western blot results of phosphorylated vs total protein ratios for three major signalling molecules of the insulin signalling pathway: ERK, AKT, and S6. Activation of all three proteins was found to be significantly elevated in KD zebrafish compared to SC (n=3, unpaired t-test **=0.0083, *=0.0498, ****<0.0001).
To manipulate the expression of grb10a, zygotes were microinjected with splice-blocking antisense oligonucleotides: e3i3 and e4i4. Two distinct splice sites, exon 3 and 4 donor splice sites (Figure 1b), were targeted in order to confirm the specificity of phenotypes accompanying oligonucleotide injection, in accordance with current guidelines (25). Disruption of the targeted splice site was confirmed through multiplexed RT-PCR amplification using a mixture of primers (F, R1 and R2) flanking the splice sites (Figure 1b). No products at all were detected for standard control oligonucleotide injected embryos and only a single product of the anticipated size was detected for embryos injected with either e3i3 or e4i4 (Figure 1c), consistent with amplification from a cDNA rather than genomic DNA template. Further, the corresponding intron was shown by sequencing to be included in the transcript for e3i3 and e4i4 treated zebrafish (data not shown).
To address the effect of grb10a KD on insulin signalling, phosphorylation of key proteins in the insulin signalling pathway was analysed by Western blotting. As shown in Figure 1d, phosphorylated (active) versus total protein ratios of AKT, ERK, and S6 were significantly elevated in grb10a KD zebrafish at 96 hpf compared to SC (p=0.0080, 0.0497, <0.0001 respectively, n=3), with S6 activation almost 5 times greater in KD zebrafish compared to SC.
Growth Trajectory and Early Life Cardiometabolic Phenotype is Significantly Impacted by Transient grb10a Perturbation
To determine the effect of grb10a KD on embryonic growth, total body length was measured in antisense oligonucleotide injected animals at various timepoints. Figure 2a shows that initially, total body length was comparable between KD and SC zebrafish (2.857 ± 0.0549 mm vs 2.826 ± 0.0962 mm, p=0.7896, n=9 and n=10 respectively). From 48 hpf, corresponding to the peak in WT grb10a expression (Figure 1a), KD zebrafish began to diverge from SC individuals. Body length was significantly different between the two groups, with KD zebrafish longer on average than SC zebrafish (3.411 ± 0.0165 mm vs 3.177 ± 0.0231 mm at 72 hpf, p<0.0001, n=46 and n=41 respectively). This phenotype was reversed in zebrafish overexpressing grb10a, with individuals injected with grb10a mRNA significantly shorter than SC counterparts (3.361 ± 0.0239 mm vs 3.505 ± 0.0339, p=0.001, n=24 and n=25 respectively), as shown in Figure 2b. Co-injection of e3i3 and grb10a RNA resulted in normalisation of growth (3.505 ± 0.0339 mm vs 3.564 ± 0.0265 mm, p=0.3792, n=25), confirming the validity of e3i3 induced grb10a KD. Moreover, grb10a KD induced by e4i4 (albeit weaker than e3i3; Figure 2c) also resulted in increased body length; while the ability of grb10a overexpression to suppress growth was shown to be dose dependent (Figure 2c). Intriguingly, by 120 hpf body length converged for both KD and OE, indicating activation of compensation to regulate growth.
2a. Mean total body length ± SEM of SC and KD zebrafish from 24 to 120 hpf. T-test significance ***=0.0006 ****<0.0001 ***=0.0001. 2b. Mean total body length ± SEM and individual data points of 96 hpf zebrafish embryos (n=25). Grb10a KD phenotype was reversed in grb10a overexpression zebrafish. Coinjection resulted in phenotype rescue. One-way ANOVA revealed KD zebrafish were significantly longer, while grb10a overexpression zebrafish were significantly smaller than SC (***=0.0001). Rescue zebrafish were of similar length to SC (ns=0.3792). 2c. Mean body length measurements relative to SC ± SEM. E3i3 and e4i4 both exhibit a propensity to elevate body length, while grb10a RNA shows a dose-dependent ability to inhibit body length. 2d. Mean heart rate ± SEM in beats per minute of SC and KD zebrafish. Heart rate was significantly lower in KD embryos compared with SC after 48 hpf (***=0.0006, ****<0.0001). 2e. Mean yolk area ± SEM of SC and KD embryos over the embryonic life stage. Following the initial 24 hours, KD zebrafish had significantly smaller yolks compared to SC, indicating yolk content was metabolised at a much higher rate (**=0.0084 ****<0.0001). 2f. Glucose Uptake-Glo ™ Assay of 96hpf KD and SC zebrafish, where higher luminescence indicates a greater accumulation of intracellular 2D6P. Luminescence was approximately 30 % greater in KD zebrafish compared to SC (***=0.0002). All SC vs KD comparisons by unpaired t-test.
Embryonic heart rate was measured to investigate the impact of grb10a KD on cardiometabolic phenotype. As shown in Figure 2d, average heart rate began to diverge between the treatment groups at 48 hpf, correlating with the WT peak in grb10a expression. While heart rate increased slightly over time in SC zebrafish, in-line with increasing body size, average KD heart rate fell. By 120 hpf, average heart rate for KD individuals was almost 50 % lower than SC individuals (61.5 ± 6.97 bpm vs 118.4 ± 2.83 bpm, p<0.0001, n=14 and n=19 respectively).
Embryonic metabolic rate was also investigated in KD and SC zebrafish by looking at yolk absorption. The reduction in yolk area over time was measured to evaluate the difference in the rate of yolk consumption between treatment groups. As shown in Figure 2e, there was no significant difference between yolk areas at 24 hpf (p=0.8185, n=10). Yolk area declined steadily from 48 hpf, and significantly more so in KD zebrafish compared to SC zebrafish for the remaining period of observation (0.298 ± 0.0055 mm2 vs 0.390 ± 0.0104 mm2, p<0.0001, n=18 and n=19 respectively). This suggested that metabolic rate was elevated in KD zebrafish compared to SC. To further interrogate insulin signalling, a Glucose Uptake-Glo™ Assay was performed to monitor 2-deoxyglucose-6-phosphate accumulation and support these findings. As shown in Figure 2f, 2D6P accumulation was significantly higher in KD zebrafish compared to SC, an increase of almost 30 % (p=0.0002, n=5). These findings are consistent with the role of grb10a as a negative regulator of the insulin signalling pathway (5).
Transient grb10a Knockdown Dysregulates Age-Related Expression in the Transcriptome
To understand whether disruption in embryonic growth trajectory could lead to lasting changes in gene expression, the transcriptomic landscape of SC and KD zebrafish was investigated over the first 30 dpf. As described in the methods, unsupervised hierarchical clustering, standard deviation filtering, and maximised projection scores were used to define a gene set with strong age-associated expression in the SC zebrafish (297 genes with a maximised projection score [PS] = 0.43).
Clear age-related associations were identified in the SC transcriptome, with genes falling into four clusters (5 dpf - 163 genes, 10 dpf-15 genes, 15 dpf – 32 genes, and 20-30 dpf – 87 genes) (Figure 3a). Clustering of these same genes was disrupted in the KD zebrafish (Figure 3b). Three of the four clusters defined in the controls mapped to the same age-associated cluster in the KD data (10 dpf – 14/15 genes, 15 dpf - 29/32 genes, and 20-30 dpf – 74/87 genes). However, genes associated with 5 dpf expression in the SC zebrafish were generally expressed at different time points in the KD zebrafish (dotted white lines, Figure 3a and Figure 3b), with a mapping of only 38/163 genes (23%). Notably, 5 genes associated with 5 dpf in SC zebrafish mapped to 20-30 dpf in the KD zebrafish. Human orthologs of these dysregulated genes included DGAT2 (fatty acid metabolism enzyme), GAMT (an enzyme involved in creatine production, required for energy storage, muscle contraction, and fatty acid oxidation), and PDIA2 (an enzyme involved in thiol-disulphide interchange reactions, particularly in the pancreas).
Hierarchically clustered heat maps of gene expression generated from an Affymetrix GeneChip™ Zebrafish Genome Array of SC (3a) and KD (3b and 3c) zebrafish RNA, taken at 5, 10, 15, 20, and 30 dpf. Expression segregates into three clusters in SC zebrafish (3a). Clustering of the same genes identified in 3a are disrupted in the KD dataset (3b). Analysing the KD dataset independently shows age-related hierarchical gene expression in the falls into two clusters (3c). 3d. Venn diagram of age associated genes. 3e. Gene set enrichment analysis, using the GO Biological Process Ontology gene list, of the age-related genes in the SC and KD datasets. The top 20 most enriched pathways with differential expression are included here, with associated normalised enrichment scores and q-values.
The existence of a pattern of gene expression associated with age in SC zebrafish which is dysregulated in KD zebrafish implies the co-ordination of an altered set of genes in KD animals. In the KD, unsupervised analysis defined a set of 119 genes (maximised projection score = 0.37) which divided into three clusters, only loosely age-associated (5-15 dpf – 37 genes, 15-30 dpf – 16 genes and 20-30 dpf – 66 genes) (Figure 3c). There was a 90 gene overlap in age-associated genes between the SC and the KD. A further 207 genes were age-associated in the SC and 29 in the KD (Figure 3d).
Gene set enrichment analysis on the age-related genes in both SC and KD zebrafish demonstrated a range of similar (Supplemental Table 1) and dissimilar pathways (Figure 3e). This analysis revealed that a range of pathways involving actin and collagen, along with extracellular structure and RNA processing, were co-ordinated in an age-related manner in KD but not SC zebrafish. Also, a set of metabolic pathways were age associated in SC zebrafish but not in KD animals. This clear dysregulation of age-related gene expression in the KD zebrafish implies transient grb10a KD induces persistent remodelling of the transcriptome.
Transient grb10a Knockdown Alters Co-Ordination Within the Transcriptome
To investigate the co-ordination of the transcriptome with age-associated gene expression, hypernetwork models were constructed for the SC and KD datasets based around genes identified by the previous unsupervised analysis. This approach can be used to model functional relationships (33). The outputs of these analyses were visualised as heatmaps where colour intensity represented the number of shared interactions between a gene pair. In SC animals, three clusters of shared interactions were defined. Cluster 1 corresponded to genes expressed at 5 dpf (161/297 genes), cluster 2 at 10 and 15 dpf (50/297 genes) and cluster 3 at 20 and 30 dpf (86/297 genes) (Figure 4a). In KD animals, however, only two clusters were identified. Genes in cluster 1 were expressed across 5, 10 and 15 dpf (31/119 genes) and genes in cluster 2 expressed at 20 and 30 dpf (65/119 genes) (Figure 4b). Notably, the large cluster of co-ordinated interactions at 5 dpf in the SC was entirely absent in the KD, instead forming a new cluster along with genes expressed at 10 and 15 dpf. The 20 and 30 dpf cluster represented 28.9% and 54.6% of the total gene set in the SC and KD zebrafish respectively, indicating a grb10a KD-induced remodelling of the transcriptome in later life.
4a. Hypernetwork analysis defined three clusters of highly connected genes in Standard Control zebrafish, corresponding to 5 dpf, 10-15 dpf, and 20-30 dpf respectively. 4b. Two clusters were defined in the KD data, corresponding to 5-15 dpf and 20-30 dpf. 4c and 4d. Violin plots of connectivity (left) and entropy (right) in the 20-30 dpf cluster. Both connectivity and entropy were elevated in the KD zebrafish compared to standard controls when controlled for random gene measurements, suggesting more diverse functions.
Differences in the co-ordination of the transcriptome in later life were quantified using two network topology parameters, connectivity and entropy. Hypernetwork connectivity quantifies the number of interactions within the transcriptome and is related to function. Entropy is a measure of information content and can be thought of as a measure of “disorder”; a decreased entropy (more order) describes a network with little crosstalk and a more discrete function (36,45). While conversely an increased entropy has less order and describes a network with increased crosstalk and pleiotropic function.
Changes in connectivity (Figure 4c) and entropy (Figure 4d) were calculated for genes identified as active at 20-30 dpf by the unsupervised and hypernetwork approaches. This was compared with entropy and connectivity calculations from a thousand iterations of random genes in order to account for random chance (Supplemental Figure 2). Genes associated with 20-30 dpf in the KD animals were found to be more highly connected and more entropic than in the SC when normalised for random effects. Additionally, this increase in connectivity is greater (1.20-fold, p < 1×10−6) and the reduction in entropy is smaller (2.0-fold, p < 1×10−6) in the KD zebrafish compared to the SC in relation to randomly selected gene sets (Supplemental Figure 2). This assessment of hypernetwork topology demonstrates the co-ordination of age-related gene expression in the SC and the dysregulation in the KD zebrafish. These data show that transient grb10 KD induces significant remodelling of the transcriptome, enduring in later life, with increased connectivity between genes and a more directed functionality.
Transcriptome-Wide Links to the Enduring Remodelling of Gene Expression Associated with Transient grb10a Knockdown
Having identified age-related gene expression we now wanted to define a wider set of associated genes which coordinate with age-related gene expression defined by calculating Galois correspondences (the one-to-one correspondence between age-related genes and their correlates).
12775 and 459 genes were defined in the Galois correspondence from the 20-30 dpf KD and SC hypernetwork clusters, respectively. 27.8-fold more genes with co-ordinated regulation at 20-30 dpf exist in the KD set than SC zebrafish. All genes in the Galois correspondence demonstrated significant rank regression with age (Supplemental Table 2). In the KD zebrafish, 7493 genes positively correlate with age (Figure 5a), while 5282 negatively correlate (Figure 5b). This was reduced to 313 positive and 146 negative co-ordinated genes in the SC. The overlap was 244 (77.9% of SC) and 60 (41.1% of SC) in the positively and negatively correlated sets, respectively. Thus, genes in the Galois correspondence with expression at 20-30 dpf in SC zebrafish demonstrated a similar pattern of expression in KD (Figure 5c). However, genes with co-ordinated expression at 20-30 dpf in the KD zebrafish were dysregulated in the SC animals (Figure 5d).
5a-5b. Venn diagrams of genes positively (left) and negatively (right) correlating with age. 5c-5d. Hierarchically clustered heat maps of gene expression of the genes identified in the Galois correspondence. Gene expression in the standard control (5c) cluster into two age related groups, whereas expression in the knockdown (5d) show significant dysregulation. 5e-5f Gene set enrichment analysis (GSEA) ranked by R-value of rank age regression in the standard control (5e) and knockdown (5f) Galois correspondence. 5g GSEA ranked by R-value of rank age regression of the cluster of genes in the white box in 5d.
Gene ontology associated with the 20-30 dpf co-ordinated gene expression clusters in both KD and SC zebrafish (GSEA ranked by R-value, top 100 pathways with weighted set cover) are outlined in Figure 5e and Figure 5f. A distinctly different set of regulated pathways was noted between the groups. The KD gene set was associated with RNA processing, a range of metabolic pathways, and cardiovascular system development. The SC gene set was associated with growth and immune pathways.
Of the genes with co-ordinated expression at 20-30 dpf in the KD zebrafish that were dysregulated in the SC animals (Figure 5d), there was a subgroup with a distinct pattern of expression at day 15, downregulation at day 20, and re-expression at day 30 (white box, Figure 5d). This subgroup of 3460 genes (Supplemental Table 3) was associated with a range of gene ontologies related to metabolism and development (Figure 5g) and is linked to the spike in growth identified in Figure 6a. Pathways included those involved in cardiovascular system development, muscle structure development, and developmental maturation.
6a. Mean total body length ± SEM of SC and KD zebrafish up to 30 dpf. Following the embryonic growth spurt, KD zebrafish experienced an additional period of rapid growth between 15 and 20 dpf (****<0.0001). 6b. Mean oxygen consumption + SEM of 30 dpf zebrafish (n=5), adjusted for body weight. KD zebrafish consumed significantly more oxygen per hour per gram of tissue than SC zebrafish (****<0.0001). 6c. qPCR results of three genes associated with cardiac performance in adult cardiac tissue, relative to β-actin. Myl7 expression was significantly elevated in KD zebrafish (****<0.0001) while nppa expression was significantly down regulated (***=0.0012) compared to SC zebrafish. There was no significant difference between the expression of pcna (p=0.3041). All comparisons by unpaired t-test.
The Impact of Transient grb10a KD Persists into Adulthood
As dysregulation of pathways associated with growth, cardiovascular system development and metabolism were identified, associated with persistent long-term changes in the transcriptome relating to grb10a KD, investigation was conducted into the differences in growth rate, oxygen consumption, and cardiac molecular phenotype in older animals.
Total body length measurements were taken up to 30 dpf. As shown in Figure 6a, growth was significantly elevated in KD zebrafish compared to SC zebrafish between 15 and 20 dpf (30 % increase, 7.493 ± 0.2726 mm vs 4.320 ± 0.1594 mm, p < 0.0001, n=10). This period of change in growth rate corresponded precisely to the cluster of dysregulated genes identified in the KD animals (Figure 5g). The growth profile of the KD zebrafish is shifted to an earlier age compared to the SC growth profile, suggesting a faster rate of maturation. The GSEA highlighted a reduction in activity in developmental maturation pathways (normalised enrichment score [NES] < 1) in the controls vs. the KDs, which was replicated in the comparison between KD and SC growth rates at approximately 20 dpf.
Juvenile (30 dpf) metabolic rate was investigated by stop-flow respirometry. As shown in Figure 6b, oxygen consumption was 7 times greater in KD zebrafish than SC (134.4 ± 15.60 µgO2h-1g-1 vs 19.0 ± 3.28 µgO2h-1g-1, p<0.0001, n=5), suggesting metabolic rate was also significantly elevated. Thus, along with significant alteration in gene expression revealed by transcriptomic analysis over juvenile phases, growth and metabolic rates were also divergent.
To address whether differences in earlier life course phases might impact adult phenotypes, the impact of transient grb10a KD on adult (1 year) cardiac health was assessed by qPCR. As shown in Figure 6c, myl7 expression (an index of muscle mass and hypertrophy) was over 20 % greater in KD zebrafish cardiac tissue compared to SC (p<0.0001, n=3), while nppa expression (cardiovascular homeostasis, associated with atrial fibrillation (46)) was down almost 40% compared to SC zebrafish (p=0.0012, n=3). This not only suggests a protective phenotype in terms of cardiac health, but also highlights the lasting impact of transient grb10a KD in adult tissue. There was no difference in the expression of pcna (proliferating cell nuclear antigen, an index of cellular proliferation), indicating there was no increase in proliferation. The greater degree of myl7 expression in the absence of an increase in pcna expression suggests an increase in hypertrophy (muscle size) rather than hyperplasia (cell number) in the cardiac tissue.
None of these cardiac phenotype marker genes were present in the Galois correspondence associated with persistent alteration of the transcriptome. However, all these genes showed dysregulated expression in the KD animals in early life (Supplemental Table 4).
Discussion
The first key finding from this study is that grb10a regulates embryonic growth in zebrafish, consistent with its role in mammalian embryogenesis (47–49). Embryonic body length was significantly elevated in KD zebrafish, and significantly repressed following grb10a overexpression, clearly demonstrating the role of grb10a as a modulator of embryonic growth. In addition, key elements in the insulin signalling pathway were significantly upregulated following grb10a KD, consistent with the role of grb10a as a negative regulator of this pathway (50,51). Transient disruption of grb10a signalling was sufficient to induce morphological, physiological, and transcriptomic changes which persisted to the late-juvenile stage, despite no further manipulation to the organism. E3i3 treated zebrafish were significantly larger during embryonic development, with significantly elevated metabolic rate, and depressed heart rate. This suggests a clear link between embryonic growth, metabolism, and cardiac phenotype, indicating that grb10a may play a fundamental role in coordinating these distinct pathways.
Additionally, metabolic rate remained elevated in the juvenile zebrafish, supporting the concept that disruptions to embryonic growth and development have a lasting impact on health and disease. Both embryonic and adult cardiac systems were impacted by grb10a KD. The ability of grb10a disruption to modify growth, metabolism, and cardiac phenotype supports a fundamental link between these three distinct aspects. As gene expression remained significantly perturbed by 30 dpf, it is clear that transient grb10a KD has persistent long-term impacts following attenuation of the morpholino, potentially indicating a lasting epigenetic remodelling. This is the only study to our knowledge to report this impact.
Transcriptomic data analysis not only confirmed a prolonged impact of transient KD, persisting after morpholino attenuation, but also revealed important gene expression trends regarding life history phase transitions in WT zebrafish. Two key transitionary phases (5 to 10 dpf and 15 to 20 dpf) were identified, with WT zebrafish undergoing a transcriptomic shift between three distinct gene clusters. These clusters correlate with the transition between larval and juvenile phases. Further investigation into these clusters will provide insight into the major factors regulating early growth and development compared to later life homeostasis.
In contrast, KD zebrafish only displayed one transitionary phase, indicating grb10a may coordinate this life-history phase transition. There was no clear transition in gene expression between 5 and 10 dpf. This suggests falling levels of grb10a (post 48 hpf peak) in WT zebrafish may induce this gene expression transition, and ablation of this grb10a gradient is sufficient to persistently remodel gene expression and to impact cardiometabolic function in later life.
Hypernetwork analysis allowed quantification of the coordination of expression within the transcriptome. There was an increase in connectivity in the hypernetwork model of KD transcriptome activity compared to controls, indicating a greater degree of interaction between age-related genes in the KD zebrafish. The hypernetwork entropy in age-related genes compared with randomly selected genes was reduced to a greater extent in the SC than the KD. A reduction in interactome entropy, as seen in SC zebrafish, is associated with more highly differentiated cell populations, corresponding to narrower biological function (36). Therefore, in KD zebrafish there will be a broader range of biological functions consistent with the increase in connectivity in the hypernetwork. In previous work looking at prepubertal growth in children we noted that greater entropy in growth hormone secretion was associated with better growth (37); these results imply that a broad range of pathways would be contributing to the increased growth hormone secretion.
The co-ordination between age-related gene expression and the wider transcriptome was assessed by measuring all genes correlated with the hypernetwork cluster (Galois correspondence). This revealed a 27.8-fold increase in co-ordinated genes in the KD zebrafish compared with SC zebrafish, and highlighted a range of associated pathways. This included cardiovascular development and metabolism in the KD, compared to immunological and growth-associated pathways in the SC. Taken together, these data indicate an ongoing impact of grb10a KD on biological functions.
Findings from this study may also provide important insight for the agri- and aquaculture industries. Embryonic heart rate was significantly depressed in KD zebrafish embryos, despite increased body length, indicating increased cardiac efficiency over SC individuals. Myl7 expression, a cardiac specific myosin light chain protein, was upregulated in the mature KD cardiac tissue, while pcna, a marker of proliferation, was not significantly different between the two groups. This suggests the difference in myl7 gene expression is due to an increase in cardiac cell size (hypertrophy) rather than cell number (hyperplasia).
In addition, the promotion of embryonic growth induced by grb10a KD may be a key alternative to growth hormone treatment for improving meat yield and production efficiency. This may be particularly useful for fisheries, as larger juvenile fish will be more capable of overwintering (52,53), and thus improve fish stocks.
Cardiac development was also linked to the persistent changes observed in the transcriptome in the KD animals. The previously described molecular cardiac phenotype markers, while not present in the clusters of genes showing persistent dysregulation, were shown to display altered age-related expression in early life. This suggests an indirect action on later life cardiac physiology.
Finally, this model of embryonic growth perturbation may yield significant insights into the propensity for early growth disruption in SGA and LGA new-borns to convey elevated risk of disease in later life. It is widely accepted that many disorders may have their origins during embryonic development (often termed the developmental origins of adult disease hypothesis (54)). Early life growth disruption and non-average for gestational age status convey elevated risk of cardiac and metabolic disorders in later life (1,55–57). While this has been observed across multiple human populations (3,58,59), the mechanisms involved are not fully understood. As grb10a modulation is sufficient to alter embryonic growth trajectory, metabolic rate, and cardiac function, this model may prove key in understanding the mechanisms involved in the developmental origins of health and disease and identifying novel avenues for prevention and treatment.
While mammalian models have been used to show the immediate impact of embryonic growth disruption, replicated in this study, their use for longitudinal study is limited, and little research has been conducted into later life impacts of early growth disruption. Due to the nature of human data collection, matched data for embryonic growth rate and later life disease risks is not possible, with measurements limited to mass and body size at birth. While this is somewhat informative, multiple distinct growth trajectories may yield similar birth weights, such as growth restriction followed by catch-up growth, or overgrowth followed by catch-down growth (55). Both catch-up and catch-down growth have been reported to correlate with increased risk of chronic health disorders (1,60,61). Furthermore, the growth profile was shifted to an earlier age in the KD zebrafish, reflecting similar growth profiles observed in cases of precocious puberty (62). This suggests the KD zebrafish may mature at a greater rate than SCs and presents an interesting model for investigating growth and maturation. This model not only facilitates growth rate measurements, but due to the short generation time is also highly suitable for longitudinal study throughout adult life.
As key physiological measurements and the transcriptomic landscape is persistently remodelled following transient grb10a disruption, this model may provide the link between embryonic development and later disease risk (63). Investigation into the metabolic and cardiac performance of the adult is the next key step.
Conclusion
This study proves transient knockdown of grb10a expression alters growth trajectory and cardiometabolic phenotype in embryonic zebrafish. This impact persists into adulthood, correlating with a measurable remodelling of the transcriptomic landscape. This provides significant evidence to suggest grb10a plays a fundamental role in the coordination of these distinct physiological pathways and could represent a promising avenue for production enhancement in an agricultural setting. The novel model of early growth disruption generated in this study, featuring easily measurable phenotypic characteristics, a short developmental window, and rapid generation time, provides much needed groundwork for thorough longitudinal study into the mechanisms underpinning the developmental origins of health and disease.
Competing Interests
The authors declare the research was conducted in the absence of any conflicts of interest.
Data availability
Transcriptomic data is available from the Gene Expression Omnibus (GSE162474).
Supplementary Figures
1a. Analysis pipeline of transcriptomic data. (i) Unsupervised hierarchical clustering was performed using a projection score to capture the most informative age-associated gene clustering in principal component analysis under a variance strip. Hypernetwork analysis was generated using the age associated genes identified in both SC and KD animals. (ii) Hypernetworks were quantified by comparing cluster structure, connectivity, and entropy. (iii) The impact of co-ordinated gene expression on the whole transcriptome was measured by identifying the Galois correspondence within the hypernetwork. 1b. A general model of a hypernetwork, shown as a three dimensional representation of genes (coloured tetrahedra) correlating with the expression of four other genes (black spheres). Shared correlations are represented by matching vertices, edges, and faces of the tetrahedra. 1c. A hypernetwork representation of the “higher order” interactions within the transcriptome. This summary of the genes with shared correlations can be considered as an incidence matrix of a multi-dimensional network.
2a. Connectivity in the experimental data was significantly greater than in the random iterative data, and greater in the KD data than SC (p<0.0001). 2b. Entropy was significantly lower in the experimental data than the random iterative data. Entropy was lower in the KD than SC, though the reduction in entropy between the experimental data and interated data was greater in the SC than the KD.
Supplementary Tables
Supplementary Table 1. Gene set enrichment analysis on age related gene expression in KD and SC zebrafish. 75212 probe ids mapped to 3733 human orthologues. Probe sets were collapsed to gene summary by averaging and group ANOVAs were performed by age groups. Q-values are reported as the false discovery rate modified p-value of the GSEA, performed in Qlucore Omics Explorer (v3.6) using Gene Ontology Biological Process pathways.
Supplementary Table 2. 12775 and 459 genes defined by the Galois correspondence from the 20-30 dpf KD and SC hypernetwork clusters respectively, together with rank regression scores and p-values. Zebrafish gene symbol and corresponding human orthologue shown.
Supplementary Table 3. A subset of 3460 genes from the Galois correspondence showing fluctuating gene expression from 15 to 30 dpf and associating with metabolism and development gene ontology pathways. Displayed as zebrafish gene symbols, corresponding human orthologues, R-statistics of age rank regression analysis, and p-values.
Supplementary Table 4. Dysregulated expression of cardiac phenotype marker genes in the KD zebrafish. False discovery rate modified p-value (q-value) shown for age group ANOVA.
Acknowledgements
This work was funded by the Biotechnology and Biological Sciences Research Council at the University of Manchester, Manchester, UK, in combination with an unrestricted Merck research grant, Darmstadt, Germany. Preliminary and the transcriptomic work was partially funded by a University of Manchester Research Institute Pump Priming Fund entitled “The quantitative comparison of the Zebrafish as a model of human development in relation to paediatric medicine”, RMS 104699.
The authors would also like to acknowledge the following people: Dr. Andrew Badrock for his training, guidance, and donation of reagents, and Jack Broadbent and Joseph Whitehead for their preliminary work in quantifying growth and metabolism in this model.
Footnotes
BLE - bridget.evans{at}postgrad.manchester.ac.uk
TG - terence.garner{at}manchester.ac.uk
CDL - c.deleonibus{at}tigem.it
OHW- wearingo{at}mcmaster.ca
HAS - holly.shiels{at}manchester.ac.uk
AFLH - adam.hurlstone{at}manchester.ac.uk
PEC - peter.clayton{at}manchester.ac.uk
AS - adam.stevens{at}manchester.ac.uk
https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE162474
References
