Single cell Calcium response analysis & modeling Michal Ronen Grischa Chandy Mary Verghese Microscopy Lab Tobias Meyer James Ferrell Data analysis group
Automatic treatment of the single cell data Outline Automatic treatment of the single cell data -Feature extraction -Discriminate analysis Model of calcium response -model -sensitivity analysis
Single cell data Feature extraction and discriminate analysis
Curve Feature Extraction Max amplitude (delta / ratio) Max slope Decrease slope storage No. of peaks Time of peaks Ca+2 (nM) Sustained (delta/ratio) basal Time (sec) Time of max slope Time of max amp Single cell trace
Discriminate analysis Experiments sets : dose experiments – compare to each other perturbations – compare RNAi transfected to control The aim – to find, automatically, the features that discriminate between the groups. The method- statistical analysis that estimates the features that have the largest separation between the groups. A problem- It does not capture small groups with different characteristics Does not handle nonlinear combinations of features. Feature 1 Feature 2 1 cell group A 1 cell group B
Feature Clustering and discriminate analysis * Cluster all data (control & experiment) * Test enrichment of each experiment group in each cluster *Apply discriminate analysis between clusters of interest expected ratio of traces in cluster enrichment = ratio of traces in cluster
Example : UDP 100nM , control & SHIP1 knockdown Data was clustered into 4 groups Clusters #2 and #4 are enriched in SHIP1 knockdown cells Discriminating features: #4 higher amplitude, higher rise-slope (10% of SHIP1, 1% control) #2 more late peaks, lower amplitude, higher sustained (35% of SHIP1, 9% control) Ca+2 (nM) Ca+2 (nM) Sample cell from cluster 4 Sample cell from cluster 2 Sample cell from clusters 1&3 Time (sec) Time (sec)
Automatic treatment of the single cell data Feature extraction is fast, easy to use & flexible All data is saved in a database (in process…) this enables using queries to retrieve whatever we want Automatic discrimination seems reliable (need more tests) takes into account all features and features combinations captures small groups of cells with different behavior
Calcium response modeling
Calcium dynamics Ca2+ agonist Calcium channel Buffer Capacitative Na/Ca Exchanger R G PLC PIP2 DAG IP3 IP3R ER agonist Ca2+ Buffer ATPase Calcium channel Capacitative entry PKC
Calcium Model R-G PLCδ IP3 Ca2+(cyt) PLCβ agonist Ca2+(ER) IP3R ER (based on Hofer et al. ,J. Neuro.,22,4850)
Calcium Model - more details (cyt) Calcium (ER) IP3R IP3
Calcium Model - more details (cyt) Calcium (ER) IP3R IP3
Calcium Model - more details (cyt) Calcium (ER) IP3R IP3
Comparing simulations to experiments - Calcium response to increased stimulation Time (sec) Ca+2 (uM) Simulations Ca+2 (nM) C5a 300nM 10nM 5nM Experiment’s single cell samples Time (sec) Ca+2 (nM) C5a 10nM Time (sec) More samples from experiments
UDP has higher sustained level than C5a Experimental data: UDP has higher sustained level than C5a Time (sec) Ca+2 (nM) Single cell samples of: UDP 10uM C5a 100nM
Sustained level could be controlled by positive feedback of calcium strength Time (sec) Ca+2 (uM) PLCδ IP3 Ca2+(cyt) PLCβ Hypothesis : perturbing this feedback will change the sustained level
Hypothesis: Sustained level could be control by negative feedback of calcium on the receptor Add feedback loop to current model & test Negative feedback strength Time (sec) Ca+2 (uM) Ca2+(cyt) R-G Only slight changes in sustained level occurred modeling addition of calcium–receptor feedback loop
Sensitivity analysis How does the model output depend upon input parameters ? Receptor level Ca+2 max amplitude (uM) G-protein level -Keep all parameters constant but one -Run model simulations -Check the correlation between changes in the parameter and model outcome OR Randomly sample the parameter space Using Latin Hypercube Sampling
Sensitivity analysis Sustained level is strongly correlated with these parameters: Ca+2 PLC feedback IP3 degradation Cor Coeff. -0.42 sustained kIP3 Cor Coeff 0.31 sustained v7 o 1 cell P-value<0.001
Sensitivity analysis Basal level is strongly correlated Cor Coeff 0.47 basal v40 Basal level is strongly correlated with these parameters: Influx Flux Out Ca+2 leak Cor Coeff. -0.66 basal k5 Cor Coeff. 0.22 basal k1 o 1 cell P-value<0.001
Future work: Further Questions: Improve the model – build a set of models better explain the measurements more reliable conclusions upon its analysis Sensitivity analysis- apply for each version of the model Further Questions: -Are the differences in the behavior of cells due to deterministic or stochastic effects?