1. The B-Cell Ligand Screen (the Final Report) and Next Steps
Madhu Natarajan, Rama Ranganathan
AFCS Annual Meeting 2004

2. The B-Cell Ligand Screen and Next Steps
Talk Outline:
Statement of the first question of the AFCS…the estimation of the level of complexity in the signaling network.
Accomplishments and conclusions from the B cell work:
Development of a general method for analysis of multivariate data.
Classification of ligand responses and estimation of interaction density.
Production of a substantial high quality dataset for the signaling community.
The future…application of the analytic concepts and methods for the next steps in the AFCS project.

3. The B-Cell Ligand Screen and Next Steps
Talk Outline:
Statement of the first question of the AFCS…the estimation of the level of complexity in the signaling network.
Accomplishments and conclusions from the B cell work:
Development of a general method for analysis of multivariate data.
Classification of ligand responses and estimation of interaction density.
Production of a substantial high quality dataset for the signaling community.
The future…application of the analytic concepts and methods for the next steps in the AFCS project.

4. The B-Cell Ligand Screen and Next Steps
Talk Outline:
Statement of the first question of the AFCS…the estimation of the level of complexity in the signaling network.
Accomplishments and conclusions from the B cell work:
Development of a general method for analysis of multivariate data.
Classification of ligand responses and estimation of interaction density.
Production of a substantial high quality dataset for the signaling community.
The future…application of the analytic concepts and methods for the next steps in the AFCS project.

5. I. The Initial Question of the AFCS.
How complex is signaling in cells? That is, what is the density of interactions between input signals in producing outputs?
Signaling Network L1 L2 L3 Ln . O1 O2 O3 Om

6. I. The Initial Question of the AFCS.
The single and double ligand screens…
Signaling Network L1 L2 L3 Ln . O1 O2 O3 Om
The ligand screen…
(1) Quantitative measurement of the similarity (or dissimilarity) of the responses to different ligands.

7. I. The Initial Question of the AFCS.
The single and double ligand screens…
Signaling Network L1 L2 L3 Ln . O1 O2 O3 Om
The ligand screen…
(1) Quantitative measurement of the similarity (or dissimilarity) of the responses to different ligands.
Quantitative evaluation of the interactions between pairs of ligand responses, and an estimation of total interaction density.

8. . II. The Analytic Methods and Results Main issues… O1 L1
Signaling Network L1 L2 L3 Ln . O1 O2 O3 Om
Calcium time points
cAMP time points .
Quantitative measurement of the similarity (or dissimilarity) of the responses to different ligands.
How can we get all the multivariate output data into general parameters that represent signaling? S-variable transformation. s
basal
Observed value

9. . II. The Analytic Methods and Results Main issues… O1 L1
Signaling Network L1 L2 L3 Ln . O1 O2 O3 Om
Calcium time points
cAMP time points .
Quantitative measurement of the similarity (or dissimilarity) of the responses to different ligands.
How can we get all the multivariate output data into general parameters that represent signaling? S-variable transformation.
Parameter reduction to appropriately represent the signaling system…as Madhu showed.

10. . II. The Analytic Methods and Results Main issues… O1 L1
Signaling Network L1 L2 L3 Ln . O1 O2 O3 Om
Calcium time points
cAMP time points .
Quantitative measurement of the similarity (or dissimilarity) of the responses to different ligands.
How can we get all the multivariate output data into general parameters that represent signaling? S-variable transformation.
Parameter reduction to appropriately represent the signaling system.
A formalism for calculating the similarity of ligand responses…the S space.

11. Quantitative measurement of similarity in ligand screen data
The Experiment Space
A highly multi-dimensional space, but one that behaves just like three-dimensional space. Each variable gets an independent dimension, and so a complete single ligand dataset is one vector in this space.

12. Quantitative measurement of similarity in ligand screen data
The Experiment Space
What can we learn from this representation?

13. Quantitative measurement of similarity in ligand screen data
The Experiment Space
What can we learn from this representation?
The response profile for each ligand is the final S vector.

14. Quantitative measurement of similarity in ligand screen data
The Experiment Space
What can we learn from this representation?
The response profile for each ligand is the final S vector.
Differences between ligand responses have a natural meaning…

15. Quantitative measurement of similarity in ligand screen data
The Experiment Space
What can we learn from this representation?
The response profile for each ligand is the final S vector.
Differences between ligand responses have a natural meaning…
DS1,2

16. Quantitative measurement of similarity in ligand screen data
The Experiment Space
What can we learn from this representation?
The response profile for each ligand is the final S vector.
Differences between ligand responses have a natural meaning…and this preserves the dimensions along which the differences occur. The cosine
DS1,2

17. Quantitative measurement of similarity in ligand screen data
The Experiment Space
What can we learn from this representation?
The response profile for each ligand is the final S vector.
Differences between ligand responses have a natural meaning…and this preserves the dimensions along which the differences occur.
How to represent dynamics?

18. The Experiment Space cAMP Ca2+ P-Proteins (min) Gene expression (hrs)
2.5
5.0 15 30
P-Proteins (min)
0.5
1.0
2.0
4.0
Gene expression (hrs)

19. The Clustered Experiment Space
cAMP
Ca2+
2.5
5.0 15 30
P-Proteins (min)
0.5
1.0
2.0
4.0
Gene expression (hrs)

20. Conclusions:
A simple transformation of raw data variables into dimensionless S variables (units of significance) permits construction of an unified experiment space of all data, regardless of source or differences in intrinsic dynamic range and signal to noise. Experiment space now contains everything we measured…Ca2+, cAMP, P-proteins, gene expression changes.

21. Conclusions:
A simple transformation of raw data variables into dimensionless S variables (units of significance) permits construction of an unified experiment space of all data, regardless of source or differences in intrinsic dynamic range and signal to noise.
The work in the past year has addressed two fundamental problems…(1) over-parameterization, the usage of many non-independent variables to represent a biological process.

22. Conclusions:
A simple transformation of raw data variables into dimensionless S variables (units of significance) permits construction of an unified experiment space of all data, regardless of source or differences in intrinsic dynamic range and signal to noise.
The work in the past year has addressed two fundamental problems…(1) over-parameterization, the usage of many non-independent variables to represent a biological process….and (2) increasing the reliability of our experiments, something particularly important in quantifying interactions between ligands.

23. Conclusions:
A simple transformation of raw data variables into dimensionless S variables (units of significance) permits construction of an unified experiment space of all data, regardless of source or differences in intrinsic dynamic range and signal to noise.
The work in the past year has helped to address two fundamental problems with this kind of analysis. One is over-parameterization, the usage of many non-independent variables to represent a biological process. The second is reliability of the measurements, a factor that becomes particularly important in evaluating interactions between ligands.
We showed that 27 out of 32 ligands applied to the B cell showed some significant response in at least some of the 134 experiment space dimensions.
The 27 positive ligands show at least 19 distinct response patterns in our experiment space.

24. I. The Initial Question of the AFCS.
The single and double ligand screens…
Signaling Network L1 L2 L3 Ln . O1 O2 O3 Om
The ligand screen…
(1) Quantitative measurement of the similarity (or dissimilarity) of the responses to different ligands.
Quantitative evaluation of the interactions between pairs of ligand responses, and an estimation of total interaction density.

25. The Initial Questions of the AFCS – Conclusions from Question 1 and Next Steps
Quantitative measurement of the “interaction” between pairs of stimuli. What experiment tests for such interaction?
Apply each ligand singly and in combination, and ask whether the response to the combined application is the additive effect of the single ligand treatments. That is…is the effect of one ligand different in the presence of another?
Signaling Network L1 L2 L3 Ln . O1 O2 O3 On

26. Does non-additivity happen in cell signaling? Yes….
J. Trimmer, Science STKE (2002), Unexpected crosstalk: Small GTPase regulation of calcium channel trafficking.
Y.Q. Xiao et al., JBC (2002), Crosstalk between ERK and p38 MAPK in TGF-b signaling.
T. Jun et al. , Science STKE (1999), Tangled webs: Evidence of crosstalk between c-Raf1 and Akt.
Y.M. Agazie et al., Am J Physiol Heart Circ Physiol, Synergistic stimulation of smooth muscle growth by ATP and insulin.
A.R. Asthagiri et al., J. Cell Sci. (2000), The role of ERK2 signals in fibronectin- and insulin-mediated DNA synthesis.
R. Laufer and J.P. Changeux, JBC (1989), Interaction between two second messenger systems in skeletal muscle.
S. Fanayan et al., JBC (2002), Interaction between IGFBP-3 and TGF-b signaling in breast cancer cells.
L. Szanto and C.R. Kahn (2000), PNAS, Leptin and insulin signaling pathways interact in a hepatic cell line.
J.M. Fredricksson et al. (2000), JBC, Interaction of b-receptor signaling and a pathway involving src in adipocytes.
A. Fatatis et al. (1994), PNAS, Synergy between VIP and a-adrenergic receptors in astroglia.
B. Gonalez et al. (2001), Endocrinology, Cooperation between LDL receptor and IGF-1 in smooth muscle proliferation.
Etc……(many many papers).
The interpretation? Non-additivity implies interaction in the signaling network.

27. Does additivity happen in cell signaling? Yes…
E.L. Greene et al. (2001) Hypertension, Additive effects of Angiotensin II and Oleic acid in muscle cell migration.
S. Shen et al. (2001) Diabetes, Additivity in PKC-d activation by insulin and IGF-1.
S. Lobert et al. (1999), Cancer Research, Additivity of Dilantin and Vinblastine effects on microtubule assembly.
S. Seraskeris et al. (2002) JCB, Additivity in a1-adrenergic receptor mediated calcium mobilization and bulk changes in intracellular calcium.
J.D. Johnson and J.P. Chang (2000), Mol. Cell. Endocrin., Additivity of different calcium pools in pituitary neurons.
J.W. Reed et al. (2000), Plant Physiol., Addtivity of several genes in controlling light-dependent growth in Arabidopsis.
G. Hiller and R. Sundler, (1999), Cell Signal., Additivity in c-PLA2 activity by several MAPKs.
Etc….many, many papers.
Additivity is taken as implying lack of interaction in the signaling network.

28. Data Analysis: The Initial Questions of the AFCS
What are the goals of the analysis?
Signaling Network L1 L2 L3 Ln . O1 O2 O3 Om
Quantitative measurement of the similarity (or dissimilarity) of the responses to different ligands.
The density of interactions between pairs of ligands. What is a good conceptual way to think about such interaction? Is the effect of one ligand different in the presence of another?
Non-additivity of inputs implies interaction in the signaling network during transduction of the two signals.
Additivity of inputs implies the of lack of such interaction.

29. Data Analysis: The Initial Questions of the AFCS
What are the goals of the analysis?
Signaling Network L1 L2 L3 Ln . O1 O2 O3 Om
Quantitative measurement of the similarity (or dissimilarity) of the responses to different ligands.
The density of interactions between pairs of ligands. What is a good conceptual way to think about such interaction? Is the effect of one ligand different in the presence of another?
Non-additivity of inputs implies interaction in the signaling network during transduction of the two signals.
Additivity of inputs implies the of lack of such interaction.
How can we use our signaling parameter (S) to represent interaction between ligands?

30. Quantitative measurement of similarity in ligand screen data
The Experiment Space
What can we learn from this representation?
The response profile for each ligand is the final S vector.

31. Quantitative measurement of interactions between ligands
Say we put on both ligands 1 and 2 together. In the case they don’t interact at all, and none of our output variables has saturated, what should we expect?

32. Quantitative measurement of interactions between ligands
Say we put on both ligands 1 and 2 together. In the case they don’t interact at all, and none of our output measurement has saturated, what should we expect?

33. Quantitative measurement of interactions between ligands
Say we put on both ligands 1 and 2 together. In the case they don’t interact at all, and none of our output measurement has saturated, what should we expect?
But what if the effect of ligand 1 changes in the background of ligand 2?

34. Quantitative measurement of interactions between ligands
Say we put on both ligands 1 and 2 together. In the case they don’t interact at all, and none of our output measurement has saturated, what should we expect?
But what if the effect of ligand 1 changes in the background of ligand 2?
The so-called “lack-of-closure” error is the interaction between the two ligands.
DDS1,2

35. Quantitative measurement of interactions between ligands
Say we put on both ligands 1 and 2 together. In the case they don’t interact at all, and none of our output measurement has saturated, what should we expect?
But what if the effect of ligand 1 changes in the background of ligand 2?
The so-called “lack-of-closure” error is the interaction between the two ligands.
This is not the same thing as the difference between two ligands (DS1,2)! This is the interaction between two ligands (DDS1,2)…the degree to which one influences the other.
DDS1,2

36. Quantitative measurement of interactions between ligands
The interaction vector preserves all the dimensions along which the two ligands interact. Thus, we can determine what experimental variables carry the interaction between two ligands.
DDS1,2

37. Examples from the double ligand screen
Let us start with the intuitive cases.
Ligand pairs that show very similar patterns in their single ligand responses seem like good candidates for non-additivity.

38. The AIG-ELC interaction
(1) The single-ligand profiles…

39. The AIG-ELC interaction
(1) The single-ligand profiles…
(2) Scaled but not reduced time series…
observed
expected

40. The AIG-ELC interaction
(1) The single-ligand profiles…
(2) Scaled but not reduced time series…
observed
(3) And the DDSAIG,ELC variables…
expected 2 3 4 5 1

41. Examples from the double ligand screen
Let us start with the intuitive cases.
Ligand pairs that show very similar patterns in their single ligand responses seem like good candidates for non-additivity.
Ligand pairs where one has no response at all seem like good candidates for additivity.

42. The LPA-TNF interaction
(1) The single-ligand profiles…
(2) Scaled but not reduced time series…
observed
expected

43. The LPA-TNF interaction
(1) The single-ligand profiles…
(2) Scaled but not reduced time series…
observed
(3) And the DDSLPA,TNF variables…
expected 2 3 4 5 1

44. Examples from the double ligand screen
Let us start with the intuitive cases.
Ligand pairs that show very similar patterns in their single ligand responses seem like good candidates for non-additivity.
Ligand pairs where one has no response at all seem like good candidates for additivity.
Now some not-so-intuitive cases.
Ligand pairs that show only weak overlap in response pattern but are still non-additive.

45. The AIG-TER interaction
(1) The single-ligand profiles…
(2) Scaled but not reduced time series…
expected
observed

46. The AIG-TER interaction
(1) The single-ligand profiles…
(2) Scaled but not reduced time series…
expected
(3) And the DDSAIG,TER variables…
observed 2 3 4 5 1

47. Examples from the double ligand screen
Let us start with the intuitive cases.
Ligand pairs that show very similar patterns in their single ligand responses seem like good candidates for non-additivity.
Ligand pairs where one has no response at all seem like good candidates for additivity.
Now some not-so-intuitive cases.
Ligand pairs that show only weak overlap in response pattern but are still non-additive.
Ligand pairs that show both positive and negative interactions.

48. The ELC-LPA interaction
(1) The single-ligand profiles…
(2) Scaled but not reduced time series…
observed
expected

49. The ELC-LPA interaction
(1) The single-ligand profiles…
(2) Scaled but not reduced time series…
(3) And the DDSELC,LPA variables…
observed
expected 2 3 4 5 1

50. Conclusions:
A new parameter of the S variable space (DDS1,2) provides a quantitative representation of the interaction between two stimuli. The calculation of the DDS1,2 vector is a complete analysis of all our data for a pair of ligands. A crutch for developing intuition…

51. Conclusions:
A new parameter of the S variable space (DDS1,2) provides a quantitative representation of the interaction between two stimuli. The calculation of the DDS1,2 vector is a complete analysis of all our data for a pair of ligands.
The analysis of interaction is now automated and could be used easily for rapid interpretation of data.

52. Conclusions:
A new parameter of the S variable space (DDS1,2) provides a quantitative representation of the interaction between two stimuli. The calculation of the DDS1,2 vector is a complete analysis of all our data for a pair of ligands.
The analysis of interaction is now automated and could be used easily for rapid interpretation of data.
Out of 171 double ligand experiments analyzed (all Ca2+ and cAMP), we find 40% that show statistically significant non-additivity.
Early results from RAW cells indicates a similar density of ligand interactions.

53. Next Steps
First, for the B cell signaling community… these data and analyses contain a large number of potentially valuable and testable hypotheses about B-cell signaling. The data are available on the web…

54. Next Steps
First, for the B cell signaling community… these data and analyses contain a large number of potentially valuable and testable hypotheses about B-cell signaling. The data are available on the web…
For the signaling community overall…

55. Next Steps For the AFCS…
(1) The analytic methods presented here are generic for any type of perturbation experiment. The RNAi experiments can be represented as just another vector in the experiment space.

56. Next Steps For the AFCS…
The analytic methods presented here are generic for any type of perturbation experiment. The RNAi experiments can be represented as just another vector in the experiment space.
How can these interaction parameters between perturbations be used to infer mechanism…the so-called “inverse problem”? What constraints are provided in modeling? Johnson, Arkin, Sternweis…

57. Acknowledgements:
Madhu Natarajan
Paul Sternweis
Elliott Ross
Mel Simon
Al Gilman

58. P-Proteins (min) Gene expression (hrs) Ca2+ cAMP 2.5 5.0 15 30 0.5 1.0
2.0
4.0

59. cAMP Ca2+ P-Proteins (min) Gene expression (hrs) 2.5 5.0 15 30 0.5 1.0
2.0
4.0
Gene expression (hrs)

60. The complete DDS matrix for the calcium dimensions
114 double ligand pairs shown