Microarray profiling emphasizes transcriptomic differences between hippocampal in vivo tissue and in vitro cultures

Abstract Primary hippocampal cell cultures are routinely used as an experimentally accessible model platform for the hippocampus and brain tissue in general. Containing multiple cell types including neurons, astrocytes and microglia in a state that can be readily analysed optically, biochemically and electrophysiologically, such cultures have been used in many in vitro studies. To what extent the in vivo environment is recapitulated in primary cultures is an on-going question. Here, we compare the transcriptomic profiles of primary hippocampal cell cultures and intact hippocampal tissue. In addition, by comparing profiles from wild type and the PrP 101LL transgenic model of prion disease, we also demonstrate that gene conservation is predominantly conserved across genetically altered lines.


Introduction
The complex interconnected structure of the mammalian brain and its anatomical location protected by the skull presents particular challenges for the study of cellular and molecular processes. In vitro cell cultures attempt to recapitulate the basic cellular environment of the brain in a more experimentally amenable context. Primary hippocampal cell cultures are regularly used as a supplement model to depict the brain's composition in a readily accessible and manipulatable arrangement. Indeed, PubMed currently cites 17 862 publications (1973-2021) associated with hippocampal cultures, and often strong conclusive inferences are drawn from these studies. Although these approaches are considered suitable model platforms, the biological relevance of cultured hippocampal cells to their in vivo counterparts is still open to question. One investigation of transcription in dorsal root ganglia and superior cervical ganglia during neurite outgrowth and regeneration, described commonalities in gene expression transcriptomics, relating to regenerating neurons between in vitro and in vivo models. 1 Similar gene expression profiles have also been detected in developing hippocampus in vivo and primary hippocampal neurons undergoing differentiation both in vivo and in vitro. [2][3][4] Conversely, genome-wide expression analysis of cell lines has indicated dramatic differences in comparison to relevant tissues of origin. 5 Remaining studies are inconclusive and describe both similarities and differences of biological processes between neural cells grown in vitro and in vivo. 6 These conflicting results suggest further studies are required to establish the full utility of cultured hippocampal cells as an in vitro model platform. Here, we compare transcriptomic profiles in acutely dissected hippocampal tissue and primary hippocampal cell cultures from both wild type (WT) animals (129/Ola) and a transgenic model of neurodegeneration based on the PrP 101LL mutation.
The 101LL model was included in this study as currently a majority of culture-based studies are being carried out to address questions about the nature of neuronal stability following specific genetic alterations/ mutations and/or neurodegenerative challenge. We sought to confirm if the degree of transcriptomic similarity holds true in murine models genetically altered with a single amino acid mutation. One such mouse line was available in our laboratory namely the PrP mutant (101LL, 129/ Ola background) 7 containing a single point mutation of the Prnp gene (proline to leucine, modelling Gerstmann-Strä ussler-Scheinker disease). This 101LL mutation is not pathological but is known to show altered susceptibility to disease associated protein misfolding. 8 Therefore, in this current study, we also sought to investigate gene expression changes in the 101LL model relative to WT. Findings suggest that RNA isolated from acutely dissected hippocampal tissue and mature in vitro primary cultures provided transcriptomic molecular fingerprints that were not comparable. This was the case for both WT and 101LL genotypes. Direct comparison between genotype (WT and mutant 101LL) revealed no (tissue) or minimal (cell) significant transcriptomic changes indicating transcript profiles were conserved across WT and 101LL genotypes.
These findings broaden our understandings of the biological relevance of cultured hippocampal neurons to their in vivo tissue counterparts and transcript changes identified here could be used to drive real progress for future therapeutic investigations using in vitro cultures.

Mouse lines
All experiments were conducted under Home Office project licence (2010-2015 PPL 60-4125: 2015-2017 PPL 70-8523) within the regulations of the Animals (Scientific Procedures) Act 1986. Study numbers A820 and A821 were approved by Roslin's Animal Welfare and Ethical Review Body. WT (129/Ola) mice were obtained from Jackson laboratories. 101LL knock in transgenic mice (129/Ola background, single point mutation, proline to leucine at codon 101 in Prnp gene) were generated inhouse using a double replacement gene targeting strategy. 7,9 WT mice were homozygous for the WT Prnp gene (101PP) and 101LL were homozygous for the P101L mutation.

Immunostaining of primary cultures
Cell media was removed from six-well plates and cells were incubated with 4% paraformaldehyde (v/v) for 15 min at room temperature. This was followed by three 5-min washes with Dulbecco's phosphate-buffered saline containing Ca 2þ and Mg 2þ (Gibco).
About 1-2 ml of ice-cold methanol was added to the cells for 10 min with incubation at À20 C followed by a 5-min incubation with 0.3% Triton-X (Sigma, v/v) at room temperature. Again, wells were washed three times in Dulbecco's phosphate-buffered saline at 5-min intervals then blocked for 1 h at room temperature using Fc Block (CD16/ 32, BioLegend). Primary antibodies were incubated overnight in 5% Goat serum (Gibco, v/v) at 4 C (concentrations listed in Table 1). Cells were washed three times in Dulbecco's phosphate-buffered saline (Gibco), and secondary antibodies diluted in 5% goat serum (v/v, Table 1) were added for 1 h at room temperature in complete darkness followed by a further three washes as above. Pre-labelled poly-L-lysine sixwell plates (Biocoat Cell Environments) were imaged using a LSM710 inverted confocal microscope (Zeiss).

Acutely dissected hippocampal tissues
Hippocampal tissues were obtained from brains of mice at postnatal Days 6-7. A non-Schedule 1 termination of each individual postnatal pup involved decapitation followed by immediate brain removal and immersion into RNAlater RNA stabilization Reagent (Qiagen). Tissues were isolated each time from three pup brains (six hippocampi) from the same litter of pups which were then combined to produce one individual sample. As this was a time sensitive protocol, brains were pooled irrespective to gender. This was replicated four times per genotype.

Cell lysis and RNA extraction
Ribonucleic acid (RNA) was extracted from both cell culture (Day 8, DIV8) and Day 6 mouse hippocampal tissue samples using the RNeasy Plus Micro Kit (Qiagen) according to manufacturer's instructions. The rationale here was that E17 (assuming gestation period of 20-21 days) harvested embryos would be cultured in vitro for 8 days to provide a more comparable developmental stage to that of day 6 tissue in vivo cells. For cell cultures, RNA extractions were always pooled in cases where more than one well was cultured from the same batch of embryos and this was counted as one sample, which was replicated four times per genotype (WT and 101LL). In total, 16 samples were generated (4 WT cell, 4 101LL cell, 4 WT tissue and 4 101LL tissue) and RNA integrity number values of 9 or above were obtained for each sample (Agilent TapeStation System) indicating high quality intact RNA was isolated.

Microarray hybridization and labelling
RNA labelling and hybridization were carried out by Edinburgh Genomics, University of Edinburgh (https://gen omics.ed.ac.uk/ Accessed 13 July 2021). For microarray, cDNA was produced using the Ambion WT expression kit (Invitrogen) and accordingly labelled using the GeneChip WT terminal labelling kit (Affymetrix). Approximately 3 mg of fragmented, biotin-labelled cDNA was hybridised to a Mouse Gene 2.1 ST array plate (Affymetrix) using the Gene Titan instrument (Affymetrix) and standard Affymetrix protocols.

Data QC and normalization
Affymetrix microarray processing produced 16 (4 WT cell, 4 101LL cell, 4 WT tissue and 4 101LL tissue) probe cell intensity data files which can be downloaded from https://doi. org/10.7488/ds/3016 Accessed 13 July 2021. Robust Multichip Average pre-processing was performed on these raw microarray intensity datasets for background subtraction, quantile normalization and summarization, using the R package 'oligo' (R package version 1.52.1). 18 A principal component analysis (PCA) plot was generated by PCA of log transformed, normalized expression data, and a clustered heatmap by calculating the Manhattan distance between sample pairs. Transcript clusters with very low expression, with no gene annotation, or with ambiguous gene annotation were subsequently removed. Differential expression was then performed using the R package 'limma' (R package version 3.44.3) 19 (Supplementary material Files 1-4; 1_diff_expr_WT_cell_vs_tissue, 2_diff_expr_101LL_cell_vs_ tissue, 3_diff_expr_cell_WT_ vs_101LL, 4_diff_expr_tissue_ WT_ vs_101LL) and gene ontology (GO) analysis was performed using the R package "topGO" (R package version 2.40.0) 20 (Supplementary material Files 5 and 6; 5_go_all_ bp_WT_cell_vs_tissue, 6_go_all_bp_101LL_cell_vs_tissue, https://doi.org/10.7488/ds/3015 Accessed 13 July 2021). Standard filtering parameters included false discovery rate (FDR) P-value <0.05. Ingenuity Pathway Analysis (IPA, Qiagen) was used to search through gene lists and determine genes involved in well documented canonical signal transduction pathways. 21 Real-time quantitative reverse transcription PCR cDNA samples (16 samples in total, 4 WT culture, 4 101LL culture, 4 WT tissue and 4 101LL tissue) at a concentration of 25 ng/ll were generated. Mastermix for real-time quantitative reverse transcription PCR (RT-qPCR) using Primerdesign was as follows; 1 ml resuspended primer mix (300 nM in a 20 ll reaction); 10 ml 2X PrecisionPLUS mastermix; 4 ml RNAse/DNAse free water. All reactions were carried out using the Stratagene Mx3005p system and SYBR green mastermix (Primerdesign/Agilent technologies). Reactions were done using 96-well PCR plates (ABgene) and optical caps (Applied Biosystems). Each sample was loaded in triplicate. To identify suitable reference/housekeeping genes, the GeNorm PCR kit (Primerdesign) was used as described in manufacture's protocol. Two cDNA samples from each representative group (WT, 101LL cultures; WT, 101LL tissue) were analysed to identify the most suitable candidate reference gene over all samples for use in normalization experiments.
Results from the GeNorm PCR kit were analysed using the Biogazelle qbase þ analysis software. Analysis results showed average expression stability of 12 reference targets ranking according to expression stability. Tyrosine 3monooxygenas/tryptophan 5-monooxygenase (Ywhaz) was stably expressed across all 16 microarray samples and therefore was selected as reference/housekeeping gene for all RT-qPCR runs.

Validation experiments
Primers were selected based on target genes of interest and included Laminin alpha 1 (Lama1), Midline 1 (Mid1), Transforming growth factor beta induced (Tgfbi), Myocyte Enhancer Factor 2 C (Mef2c) and Transthyretin (Ttr). Relative changes in gene expression were calculated using the Delta Delta Ct (DDC T ) method. 22,23 IMARIS software analysis of immunolabelled hippocampal culture images IMARIS software (Bitplane) allowed for data visualization and analysis of confocal microscopy datasets, in the format 'czi'. For each image or channel within an image the intensities of all voxels based on fluorescent signal were analysed using default standard IMARIS formulas that calculated mean, standard deviation and sum intensities (intensities do not have any units) and therefore these values were used for relative comparison of targets of interest across all comparative images.

Statistical analysis
All graphs and statistics were generated in GraphPad Prism 9. Normality and Lognormality (Anderson-Darling, D'Agostino-Pearson and Shapiro-Wilk) tests were performed prior to any statistical testing. If data sets passed the normality test, a one-way ANOVA with Tukey's/ Sidaks post-hoc was carried out. When data sets did not pass the normality test a non-parametric Mann-Whitney test was carried out. For statistical tests, P < 0.05 was used for significance. All ANOVA tests were presented with F and P values for main effects.
Significant effects between groups, identified by posthoc analysis, were displayed visually on graphs and recorded in text as P-values. All data were plotted as means with 95% CI for normal distribution and medians with 95% CI for non-parametric data.

Compliance with ethical standards
All applicable international, national and institutional guidelines for the care and use of animals were followed. All procedures performed in studies involving animals were in accordance with the ethical standards of the institution at which the studies were conducted. The article does not contain any studies with human participants performed by any of the authors.
101LL cultures were comparable regarding cellular profile, immunolabelled images of MAP2/Synapsin1/PSD-95 DIV8 cell cultures were processed using IMARIS software. Data intensity comparisons established from fluorescent signal for each target protein showed no significant differences between WT and 101LL cell cultures indicating both had similar cellular profiles (Fig.  1M). This also reinforced reproducibility of the culture method used across genotype. Overall, these cultures were similar to those reported in the literature 17,24,25 and were appropriate for use in the comparative molecular fingerprinting experiments proposed here.

Microarray profiling establishes transcriptomic differences between hippocampal in vivo tissue and in vitro cell cultures comparisons
The microarray platform was employed here to compare gene expression in both in vitro hippocampal cell culture and in vivo hippocampal tissue RNA extracts to establish to what extent these contexts were or were not comparable. A PCA plot (first and second components) generated from microarray analysis of all 16 in vivo and in vitro samples (4 WT cell, 4 101LL cell, 4 WT tissue and 4 101LL tissue) was used to visualize patterns associated with these datasets (Fig. 2A).
The visualization plot generated confirmed a clear pattern and separation was evident between cell (orange) and tissue (blue) groups of arrays analysed by principal component. This was independent of genotype.
These observations were further supported by clustering all 16 samples on a heatmap (Fig. 2B), where again differences in gene expression patterns were evident between cell and tissue platforms. To ascertain the magnitude of gene changes between in vivo and in vitro, standard filtering of datasets was carried out using an FDR P-value of <0.05. To visualize these changes, scatter graphs were produced accordingly. Scatter plots of both WT cell versus tissue and 101LL cell versus tissue (Fig. 2C and D) clearly show numerous differentially expressed genes (DEGs) were detected using standard filtering parameters (FDR P-value <0.05). Here each dot/triangle resembles a single gene, genes above the dotted horizonal line were significantly changed. For WT cell versus tissue 5199 DEGs (2706 down and 2493 up) were detected (Fig. 2C). When applying additional fold change parameters of two (donated by dotted vertical lines) 830 DEGs were detected of which 682 were downregulated and 148 were upregulated. In comparison, 101LL cell versus tissue produced similar results with 4677 DEGs identified (2527 down and 2150 up) at FDR P-value <0.05 (Fig. 2D). When a fold change of two was included, this number was decreased to 856 DEGs (660 down and 196 up). As highlighted on both scatter graphs similar genes were changing across both genotypes, including downregulation of Adgrl4, Atp13a5, Car4 and upregulation of Bace2, Car14, Ephx1.
Comparison of 101LL versus WT tissue identified no DEGs using P-value <0.05 indicating the single point mutation in the Prnp gene did not change baseline transcript profile in vivo. Whilst comparing 101LL versus WT cell, 5 DEGs were identified, and all were upregulated in the 101LL genotype. One of these genes namely Midline 1 (Mid1), is associated with microtubule stabilization 26 suggesting that neurons may be less stable in the 101LL genotype thus, increasing expression of Mid1 may be protective. Indeed, it is known that these transgenic animals show altered susceptibility to disease and this may well explanation observations here. 8 Collectively, it is evident here that baseline transcriptomic profiles are analogous between WT murine models and genetically altered models with a single amino acid mutation.

IPA highlights multiple affected pathways
To investigate biological functions associated with DEGs identified and the pathways they influence, significantly filtered datasets from both WT and 101LL cell versus tissue were analysed using topGO and IPA. Biological GO terms identified using topGO were numerous (Supplementary material File 5 and 6, https://doi.org/10. 7488/ds/3015 Accessed 13 July 2021) and were comparable across WT cell versus tissue and 101LL cell versus tissue datasets.
Briefly, top biological processes identified in vitro and similar to both datasets included the following, adenylate cyclase-activating G protein-coupled receptor signalling pathway, MAPK cascade, learning or memory, long-term synaptic potentiation and calcium ion transport which were all downregulated. Biological processes upregulated in vitro included cholesterol biosynthetic process, regulation of cell growth, forebrain neuron development, regulation of Wnt signalling pathway and endoplasmic reticulum unfolded protein response. IPA also provided an alternative means of data analysis and interpretation. This software is built on a very comprehensive and manually curated knowledge database and therefore provides unique capabilities to identify the most significant pathways, whether activated or inhibited from our experimental data. Evidence from Fig. 3A (FDR P-value <0.05) shows top canonical pathways identified in IPA are conserved between genotypes and are predominantly inhibited/downregulated in vitro. Canonical pathways identified here complement topGO results where neurotransmission, cell signalling, and memory were all inhibited in vitro. Activated pathways are associated with cholesterol synthesis Wnt signalling and ER unfolded protein response. To gain further insights into gene expression changes in our datasets we included a heatmap of the top 20 upstream regulators (Fig. 3B). Here, we show genes identified were consistently expressed across both WT cell versus tissue and 101LL cell versus tissue.
The most significant canonical pathway identified was CREB signalling in neurons and this was shown to be inhibited in vitro (Fig. 3A). The CREB as a nuclear transcription factor binds to CRE (cAMP response element, which is also shown the be inhibited in vitro), and regulates transcription activity of its downstream substrates. This in turn regulates neuronal processes, including metabolism and survival. Several gene changes were identified in this cascade and as shown in Fig. 3C were all predominantly downregulated.

Validation experiments
Microarray expression results ( Table 2) were validated using RT-qPCR based on significant DEGs identified (Lama1, Mid1 and Tgfbi) in the 101LL versus WT cell cohort (Fig. 4A). For completeness and robustness additional genes that were either downregulated (Mef2c) or not changed between comparisons (Ttr) were also included in this validation. As evident from Fig. 4B-F similar expression trends between all comparisons were found using RT-qPCR thus validating microarray data, and confirming results presented here were accurate.

Discussion
This study has attempted to address the lack of data detailing the molecular composition of cell culture platforms used to model and infer upon their intact in vivo neuronal counterparts. Studies by others have attempted similar investigations, however, direct comparisons between identical genotypes from both in vivo and in vitro contexts was never carried out, instead separate studies were combined to study transcriptomic changes between both. For example, one study focussed on murine gene expression in developing hippocampus in vivo, 2 and a second study on expression profiling of primary hippocampal neurons undergoing differentiation in vitro. 3,4 Both studies were then combined for subsequent comparisons. Results showed in vitro and in vivo expression profiles were similar. These findings contradict the results presented here however their study was comparing primary cultures obtained from CD1 outbred mice which have more genetic diversity 4 with hippocampal tissue from C57BL/6 inbred mice which are almost genetically identical. 2 Therefore, these studies did not represent an accurate direct comparison of in vitro and in vivo platforms. Another study using a similar approach (combining two separate studies) described both similarities and differences of biological processes between neural cells grown in vitro and in vivo. 6 However, these results were based on rat neural cells obtained from commercial sources (in vitro) compared to mouse acutely purified neural cells (in vivo), implying these data involved studying cross-species transcriptomic comparisons. Thus, although in vitro and in vivo comparisons have been carried out previously, the studies were restrictive. Studies presented in both the introduction and here in the discussion suggest complex in vivo tissues consisting of multiple cell types have similar gene expression profiles to neuronal cultures. These findings are questionable as differences in in vitro cultures would be expected due to the simplicity of the platform where many of these transcriptional changes would be driven by the emission of cell types present in vivo. In this study, a direct comparison between in vivo/in vitro models and genotype was done and as expected changes in gene expression between organized in vivo tissues were detected. However, for a more accurate transcriptional profile comparison, singlecell RNA-seq comparing gene expression of individual cell types such as neurons between both in vivo and in vitro platforms could be carried out and this was a limitation of our study here which only used bulk RNA-seq.
Gene expression changes presented here indicated that primary hippocampal cultures and acutely dissected hippocampal tissues were not comparable and numerous biological pathways were perturbed in vitro. Pathways relating to memory again would be expected to be inhibited in cultures as shown here however many other pathways relating to a broad range of biological pathways were also disturbed including neurotransmission, cell signalling, cholesterol synthesis, Wnt signalling and ER unfolded protein response. Again, studies utilizing in vitro platforms to study such pathways should be cautious in their interpretations of results.
Interestingly, gene expression was not altered between WT and mutant genotype (apart from 5 DEGs in cell) indicating a single amino acid mutation may not alter detectable transcriptomic changes between transgenic and WT models. Thus, by comparing profiles from WT and the PrP 101LL transgenic model of prion disease, we demonstrate that gene conservation is predominantly conserved across genetically altered lines. It is possible by using other technologies further transcriptional changes could be detected as microarray hybridization is restricted to predefined transcripts/genes and this is one limitation of the study presented here. RNA-sequencing, for example, could profile the entire transcriptome of these models and therefore could be used to detect more subtle transcriptional changes.
In conclusion, we have shown adequate evidence that primary hippocampal cultures are significantly different to their in vivo tissue counterparts at a transcriptional level, and one should be cautious when planning and interpreting data from primary cultures. This study also provides a unique transcriptome resource and a list of canonical pathways that are significantly altered in vitro. These insights should help in future experimental planning or to re-access previously published data based on neuronal cell culture models.

Supplementary material
Supplementary material is available at Brain Communications online.