in A673V in A673T
Superoxide dismutase P04179 DOWN DOWN Destroys superoxide anion radicals X X X X
Serum albumin P02768 DOWN DOWN promotes neuronal survival X X
Flotillin-2 Q14254DOWN DOWN
negative regulation of amyloid precursor
protein catabolic processX X X
Tubulin alpha-8 chain Q9NY65UP DOWN
Translocation of GLUT4 to the plasma
membrane, GAP junction assemblyX X
Transmembrane protein 35 Q53FP2UP DOWN
interact with NGFR and modulate neurite
outgrowthX X
AP-3 complex subunit delta-1 O14617 UP N.R. neurotransmitter receptor transport X X X
Chloride transport protein P51797 UP N.R. voltage-gated chloride channel activity X X
Pre-mRNA-processing factor 40 homolog
A
O75915UP N.R.
regulation of cell morphology and
cytoskeletal organizationX X X X
Protein arginine N-methyltransferase 1 Q99873 UP N.R. transcriptional activity, neurite outgrowth X X X X X X
Emerin P50402UP N.R.
Stabilizes and promotes the formation of a
nuclear actin cortical networkX X X X X X
Eukaryotic translation initiation factor 5A-
1
P63241UP N.R.
mRNA-binding protein, brain developmentX X X X X
Glutathione synthetase P48637 DOWN N.R. nervous system development X X X
Neurosecretory protein VGF O15240UP N.R.
cell-cell interactions or in synatogenesis,
neuropeptide activityX
Nuclear pore membrane glycoprotein
210
Q8TEM1DOWN UP
nuclear pore assembly X X
Sialic acid synthase Q9NR45 UP N.R. carbohydrate biosynthetic process X
Actin-related protein 2/3 complex
subunit 2
O15144UP N.R.
Actin polymerization X X
GPI inositol-deacylase Q75T13DOWN N.R.
forebrain regionalization, protein transportX X
cAMP-dependent protein kinase catalytic
subunit beta
P22694UP UP
activation of protein kinase A activityX X
Signal recognition particle 19 kDa protein UP UP
Signal-recognition-particle assembly X
Dual specificity mitogen-activated
protein kinase kinase 2
P36507;Q0275
0UP UP
activation of MAPK activityX X X
Carboxypeptidase E P16870UP DOWN
metabolic processes, neuropeptide
signaling pathway X X
Leukocyte surface antigen CD47 Q08722 DOWN N.R. cell adhesion , integrins X X
Cellular retinoic acid-binding protein 1/2 P29762,
P29763UP N.R.
cellular developmentX X
High mobility group protein HMG-I/HMG-
Y
P17096DOWN N.R.
chromatin bindingX X
Nuclear ubiquitous casein and cyclin-
dependent kinase substrate 1
Q9H1E3UP N.R.
chromatin binding, regulation of insulin
receptor signaling pathwayX X
Prefoldin subunit 5 Q99471 UP N.R. chaperone activity X X
Small nuclear ribonucleoprotein Sm
D1/D2
P62316,
P62314UP DOWN
mRNA splicing X X
Elongation factor Tu, mitochondrial P49411 UP N.R. protein biosynthesis X
Eukaryotic translation initiation factor 3
subunit G
O75821UP N.R.
initiation of protein synthesis X
U4/U6 small nuclear ribonucleoprotein
Prp3
O43395UP N.R.
pre-mRNA splicingX
Name ID RoleSynaptic
plasticity
Transcription regulatedMembrane
organization
Gene splicing,
expression
Nuclear
transport
Axonal
transport,
cytoskeleton
Metabolism
(MAPK, PKA,
Wnt)
Oxidative
stress /
apoptosis
0
20
40
60
80
100
120
iCellDopa Neurons iCellGluta Neurons iCell Neurons
Functional Phenotypic SimilarityFrontal Cortex
Midbrain/Cortex
Hippocampus
Spinal Cord
Poisson
iia
%s
ml
r
classifier
FC
Hc
MB
+F
C
SC
Pois
son
FC 81 4.2 14 0.2 0.2
Hc 2.6 94 1 2.4 0
MB+FC 13 2 82 2.8 0
SC 1.2 1.6 2 94 0.2
Poisson 0 0 0 0 100
Genotype/Phenotype correlation in human iPSC-derived neuronal networks and in vitro HTS disease modeling with micro-electrode arrays coupled with transcriptome analyses.
*1 2 1Benjamin M. Bader , Maren Depke , Monica Segura-Castell , Maria 1 1 2 1Winkler , Konstantin Jügelt , Frank Schmidt , Olaf H.-U. Schröder
1 NeuroProof GmbH, Rostock, Germany; contact: [email protected] University Medicine Greifswald, Dept. of Functional Genomics, Greifswald, Germany
IntroductionTo decrease attrition rates is one of the major challenges of drug discov-
ery. One strategy is the development of more predictive pre-clinical in
vitro models. Human induced pluripotent stem cell-derived (hiPSC)
neuronal cultures promise higher physiological relevance and thus, better
translation to the in vivo situation. In this context genetic cell models have
been designed or produced from patient biopsies. However, one of the
biggest concerns is their physiological relevance needed for disease mod-
eling.
Our aims were a) to identify a correlation between known disease geno-
types (e.g. of familiar Alzheimer's disease) and their functional in vitro phe-
notype and b) to better understand these human iPSC neuronal in vitro
models by comparing them with primary mouse neuronal cell cultures.
Methods ConclusionsPrimary culture: Primary mouse tissue cultures from embryos (NMRI) were
cultured on MEAs for 4 weeks.
hiPSC culture: We cultured human iPSC Neurons (Cellular Dynamics
International, USA) on multi-well MEAs (Axion Biosystems) for up to 4 weeks.
MEA recording: MEAs from Axion Biosystems, USA were recorded on the
MAESTRO recording station. Per data point 60 minutes were recorded at
37°C und stable pH condition.
MEA Data analysis: Acquired spike train data was analysed multi-
parametrically yielding more than 200 parameters (NPWaveX software).
Classification analyses were performed using PatternExpert software.
Transcriptomic analyses: RNA quality control with Agilent Bianalyzer, whole
genome Affymetrix GeneChip Human Transcriptome Array 2.0, Gene
Ontology analysis.
By comparing human functional phenotypes with those from different specific
mouse brain regions we show that human neuronal cell exhibit specific and
reproducible phenotypic similarity profiles (e.g. midbrain or hippocampus-
like). Genetic manipulation affects these activity patterns thereby producing
specific and significant functional phenotypes.
For APP-mutated cells our transcriptome results showed regulation of more
than 100 genes including genes involved in membrane plasticity, axonal trans-
port, gene splicing, cell cycle, glutamate receptor signaling and neuro-
protection.
In conclusion, we show that hiPSC neurons are able to produce meaningful
functional in vitro phenotypes, that these phenotypes can be changed by dis-
ease-associated modulation and we provide first data for a genotype/
transcriptome/ phenotype correlation.
ResultsBrain Region-Specific Cell Cultures with Unique Network Activity Patterns.
Brain region-specific neuronal cell cultures from mice and a random spike train. Network spike train patterns of brain-region specific primary cell cultures derived from embryonic mouse tissue of frontal cortex (FC), spinal cord (SC, with dorsal root ganglia), hippocampus (Hc), and midbrain co-cultured with frontal cortex ( Mb+FC). Plotted are 60 s of 25 neurons of spontaneous network activity at 28 days in vitro.
\
\ \ \
\\
\
0
1
2
3
4
5
6
7
8
7 div 14 div 21 div
Mean
S
EM
[1/s
]
Mean Spike Rate - HC
0
1
2
3
4
5
6
7
8
6 div 14 div 21 div
Mean
SE
M [1/s
]
Mean Spike Rate - FC
\\
\
Frontal Cortex Hippocampus
Spinal Cord
Midbrain+Frontal Cortex
Poisson (random spike train)
Example MEA spike trains of iNeuron cell lines after 3-21 daysin vitro after thawing. Cultured on 48-well MEAs with 16 electrodes each. A high level of synchronization is shown by strong populations bursts for iCell Gluta and iCell Dopa Neurons. On 48 well MEAs a unit separation was performed yielding up to 32 active units on 16 electrodes per well.
Cross validation shows that activity patterns are unique (high % self-recognition) and thus, also highly reproducible. Average values of 5 classification rounds with 100 data sets each using the combination of more than 200 parameters.
Comparing phenotypes from human iPSC-neurons to primary neurons.
Human iPSC-derived Neuronal Activity Patterns.
Classification of activity patterns from three different human iPSC derived neurons during 1-4 weeks in culture shows a cell line specific phenotype (=fingerprint showing a distribution of similarities compared to primary mouse neuronal cultures). ICell Dopa Neurons show the highest similarity to ventral midbrain cultures mixed with cortex. ICell Gluta Neurons are most similar to primary hippocampus cultures and iCell Neurons are most similar to the poisson-generated in silico spike train which underlines their lower complexity. MyCellNeurons are differentiated with the iCellN protocol.
These results highlight the importance to compare human with known reference data and show a) differences in quality of human iPSC neurons and b) physiological relevance.
ICell Gluta Neurons at 21 div oICell Neurons at 7 div
240 sec
ICell Dopa Neurons at 21 div
300 sec
300 sec 300 sec
Phenotypic Screening with MEA-Neurochips
Synch
ron
zatioi
n
Full disinhibition (with bicuculline, strychnine, NBQX)
Native activity 30 s
Oscillation
2 Burst Structuree.g. number, frequency and ISI of spikes in bursts; burst duration, amplitude, area, plateau position, plateau duration
1 General Activity e.g. spike rate, burst rate, burst period, percent of spikes in burst
Read out: Extracellular action potentials on a single neuron and network activity level Spatio-temporal activity changes as well as synchronicity and oscillation in time scales
of spikes and bursts
!
!
Each specific spike train is described by 200 parameters in 4 categories:
3 OscillationVariation over time as an indicator for the strength of the oscillation; in addition e.g. Gabor function parameters fitted to autocorrelograms
4 SynchronizationVariation within the network as an indicator for the strength of the synchronization; in addition e.g. simplex synchronization, percent of units in synchronized burst
Multiparametric Characterization of Neuronal Network ActivityNeuronal
Cell Culture Multichannel Recording
Multiparametric
Data Analysis
Pattern
Recognition
Primary murine cell culture:- Frontal Cortex- Hippocampus- Midbrain/Cortex- Spinal Cord/DRGNeuronal human Stem Cells
Network spike trains and single neuron action potential
Over 200 descriptors at baseline and drug treatment- General activity- Synchronization- Oscillation- Burst structure
Data base with functional fingerprints of over 100 basic and clinically compounds
MAESTRO Recording System
Axion Maestro MEA recording Station Neuronal network on electrode field
NeuroProof Technology
Burst Amplitude
Burst Duration
Burst Period
Burst IBI
Burst Plateau
Burst Area
Burst ISI
Burst Duration
burst
Controlactivity
Patient activityclassification
!
Comparison of activity frompatient and control iPSC‘s
48 well MEA with 16 electrodes per well
Machine Learning/classification paradigm
Classification of spike train-derived activity pattern
into discrete classes enables phenotype
identification
300 sec
MyCell Control Neurons at 7 div
In vitro Alzheimer’s Disease modeling: functional biomarker for APP mutation in human neurons.
*
0
50
100
150
200
250
300
350
400
450
7-8 div
culture time
Me
an
±S
EM
***
0
1
2
3
4
5
6
7
7-8 div
culture time
Me
an
±S
EM
***
0
2
4
6
8
10
12
14
16
18
7-8 div
culture time
Me
an
±S
EM
***
*
0
0.1
0.2
0.3
0.4
0.5
7-8 div
culture time
Me
an
±S
EM
[%]
**
*
0
100
200
300
400
500
600
7-8 div
culture time
Me
an
±S
EM
***
0
1
2
3
4
5
6
7
8
7-8 div
culture time
Me
an
±S
EM
0
0.2
0.4
0.6
0.8
1
1.2
7-8 div
culture time
Me
an
±S
EM
**
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
7-8 div
culture timeM
ea
n±
SE
M
***
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
7-8 div
culture time
Me
an
±S
EM
***
Spike rate
0
0.2
0.4
0.6
0.8
1
1.2
1.4
7-8 div
culture time
Me
an
±S
EM
[%]
Spike contrast
0
0.05
0.1
0.15
0.2
0.25
0.3
7-8 div
culture time
Me
an
±S
EM
[%]
Burst rate
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
7-8 div
culture timeM
ea
n±
SE
M
General activity Burst structure
Oscillation Synchronicity
Isogenic control Protective APP673T Causative APP673V
Functional activity of control neurons (MyCell control, isogenic to mutants) and genetically engineered iPSC-derived neurons with mutations in the APP gene (amyloid precursor protein) leading to a causative genetic variant linked to Alzheimer’s Disease (APP673V) and a protective form (APP673T) not leading to the disease. Shown are 12 functional parameters indicating the phenotypic difference as a result of the APP673V mutation. Interestingly, these are not part of the general activity parameter category but seen in burst structure, oscillation and synchronicity.
Isogenic control Causative APP673V mutationSpike train examples for spontaneous activity of APP673V mutant and isogenic control at 8 days in v i t r o s h o w d i f fe re nt patterns: e.g. Shown by longer bursts in APP673V. 300 sec 300 sec
300 sec
Protective APP673V mutation
Effect score ± SEM
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
control causative
APP673V
protective
APP673T
*** ****
combination ofall parameters
*
Position ofa potential
new AD drug?
These results underline the need for multi-variate spike train analysis with focus on neuron-to-neuron communication. This approach provides a means to screen test compounds to rescue this disease-
associated functional “biomarker” towards control conditions.
“Effect Score” calculation: Projection of up to 204 parameters into a single parameter allows comparing multivariate data statistically and ranking of rescue efficacies based on the complete functional finger print. We calculate an optimized linear combination of selected features for an optimal separation of control effects from those of APP673V and APP673T. Control is set to “0”, APP673V is set to “1”. The calculated Z-factor (describing the effect size) is optimized to find the best discrimination between the two groups.
Transcriptomic Analysis of APP673V, APP673T and control cells after MEA recording.
MEA recording
RNA quality control using Agilent 2100 Bioanalyzer
Affymetrix whole human genome chip analysis
Gene Ontology annotation
Selection of regulated genes:
Affimetrix analysis of MEA-recorded human neurons. Cells were harvested after MEA recording and RNA was isolated, quality controlled using Agilent 2100 Bioanalyzer. Whole genome Affymetrix GeneChip Human Transcriptome Arrays were used to investigate transcriptional differences between the isogenic lines. Gene Ontology analyses revealed more than 100 genes regulated with at least 1.5 fold above levels measured in isogenic control lines. The table shows a selection of genes which were expressed differently between mutants and control (at least two-fold).
Multiple genes involved in synaptic plasticity, membrane organization, axonal transport were affected and thus, are directly involved in synaptic functional which is one explanation for the functional phenotypic difference observed in the MEA recordings. Moreover, genes involved in cellular signaling (MAPK, PKA, canonical Wnt), oxidative stress response are differently expressed. Genes involved in gene splicing, protein expression and nuclear transport were upregulated in causative APP673V mutant cells.
RNA isolation
mage fro
thm
risherc
mI
m
ee
of.
o
Supported by: