Slides from my thesis defense. I discuss why we need more databases in neuroscience and talk about neuroelectro.org, a resource I've built on neuron types and their properties. I also talk about integrating neuron physiology information with gene expression information

1. Understanding the form and
function of neuron diversity
Shreejoy Tripathy
Carnegie Mellon University
Email: stripat3@gmail.com
Twitter: @neuronJoy
1

2. Thesis overview
1. How are neurons
within the same
type different?
2
Tripathy et al 2013, PNAS

3. Thesis overview
1. How are neurons
within a type
different?
2. How are neuron
types throughout
the brain different?
3
Tripathy et al, in preparation

4. Thesis overview
1. How are neurons
within a type
different?
2. How are neuron
types throughout
the brain different?
3. How can we best
utilize existing data
and knowledge?
4
?

5. Outline
• Background on neuron electrophysiology
• The role of neuron variability
• NeuroElectro: a window to the world’s
neurophysiology data
• Analyzing the electrophysiological diversity of
neurons throughout the brain
• Extending NeuroElectro and future directions
5

6. Ways of looking at neurons:
Neuron morphology
Ramon y Cajal, 1905
6

7. Neuron electrophysiology
Membrane
voltage
Time
Current
injection
7

8. Ion channel basis of electrophysiology
8

9. Summarizing neuronal
electrophysiological properties
9
Membrane
voltage
Time

10. Outline
• Background on neuron electrophysiology
• The role of within-type neuron variability
– How are olfactory bulb mitral cells different from
one-another?
– What is the computational role of mitral cell
differences?
10

11. Mitral cell variability
11
Krishnan Padmanabhan
Now at Salk Institute
Olfactory bulb mitral cells
(named because their shape looks like
a bishop’s hat, or mitre)
• Individual mitral cells
are biophysically
variable from one
another
Padmanabhan and Urban 2010, Nat Neuro

12. Modeling and explaining the
significance of mitral cell variability
• Captured the
variability across
mitral cells using
statistical models
– Models allowed for
precisely quantifying
neuron differences
12
Tripathy et al 2013, PNAS

13. Modeling and explaining the
significance of mitral cell variability
• Captured the
variability across mitral
cells using statistical
models
• Mitral cell populations
optimally encode
information when at
an intermediate level
of variability
13
Tripathy et al 2013, PNAS

14. Outline
• Background on neuron electrophysiology
• The role of neuron variability
• NeuroElectro: a window to the world’s
neurophysiology data
– Why do we need a neurophysiology database?
– Live demo of web interface at neuroelectro.org
– Brief methods explanation
14

15. Extending neuron comparisons to
other neuron types
• What makes mitral cells
unique?
• Could I extend my results
on mitral cells to other
neuron types?
– “Is a mitral cell more like a
CA1 pyramidal cell or a
cortical basket cell?”
15

16. Extending neuron comparisons to
other neuron types
• What makes mitral cells
unique?
• Could I extend my results
on mitral cells to other
neuron types?
– “Is a mitral cell more like a
CA1 pyramidal cell or a
cortical basket cell?”
16
Pubmed searches
Recording new data

17. Extending neuron comparisons to
other neuron types
• What makes mitral cells
unique?
• Could I extend my results
on mitral cells to other
neuron types?
– “Is a mitral cell more like a
CA1 pyramidal cell or a
cortical basket cell?”
17
Pubmed searches
Recording new data
The lack of a database on neuron
types and their properties leads to:
• Increased difficulty in comparing
neurons
• Overall slower progress and more
narrow-minded focus in the field

18. Availability of useful databases in
genetics
18
Entrez
tools
CCCATTGCGCCAAGCCCGTT…
CCCATAGGGCCAAGTTCGTT…
97%; human Kv1.1 (KCNA1)
95%; mouse Kv1.1 (Kcna1)
Genetic deficit in gene coding for
Kv1.1 potassium channel
CCCATTGCGCCTAGGGCGTT…

19. NeuroElectro: text-mining neuron
properties from the existing literature
Tripathy et al, in preparation
inspired by Aaron Swartz,
IBM’s Watson, neurosynth.org
19

20. NeuroElectro web interface
• [demonstration of http://neuroelectro.org
web interface]
20
Built with Rick Gerkin
Now at Arizona State

21. Semi-automated literature-mining
overview
• Simple algorithms that use
simple text searching to
identify:
– Biophysical properties (in
normotypic conditions)
– Neuron types (from
NeuroLex.org)
– Biophysical data values
– Methodological details
• Text-mined data is then
checked by experts
>+
21
Tripathy et al, in preparation

22. NeuroElectro: a window to the world’s
neurophysiology data
• NeuroElectro: a database of neuron types and their
biophysical properties in normotypic conditions
– Built through literature text-mining + manual curation
– Currently 100 neuron types, >300 articles
• Anyone can access, visualize, and download this data at
neuroelectro.org
• Currently working with computational modelers to
provide parameters for neuron models
• Future efforts will integrate raw data, data not from
normotypic conditions including animal models of
disease
22

23. Outline
• Background on neuron electrophysiology
• The role of neuron variability
• NeuroElectro: a window to the world’s
neurophysiology data
• Analyzing the electrophysiological diversity of
neurons throughout the brain
– How do we reconcile data collected under different
experimental conditions?
– What new things can we learn about functional
relationships between different neuron types?
23

24. Example electrophysiological data
24

25. How to deal with differences in
experimental conditions?
25

26. How to deal with differences in
experimental conditions?
• A tale of (representative) 2 labs
Urban Lab Barth Lab
“Slices were continuously superfused with
oxygenated Ringer’s solution warmed to 37°C”
-Burton et al 2012
“Slices were maintained and whole-cell
recordings were performed at room temperature”
-Wen et al 2013
26
Stated reason: “more physiological,
more realistic results”
Stated reason: “healthier neurons”

27. How to deal with differences in
experimental conditions?
27
• Experimental condition differences are
relevant to all experimental disciplines
• Addressing effect of differences in
experimental conditions is generally quite
difficult
– “I only believe my data and noone elses”

28. Example experimental conditions
28

29. Example experimental conditions
29

30. How to deal with differences in
experimental conditions?
• Idea: use statistics to model the influence of
experimental conditions (“metadata”) on
electrophysiological measurements (“data”)
30
Spruston et al 1992
Zhu et al 200
Electrode type

31. Influence of specific experimental
metadata
31

32. Explaining measurement variance with
experimental metadata
32

33. Analyzing neuronal
electrophysiological diversity
• How do we reconcile data collected under different
experimental conditions?
– Use statistics to learn the relationship between
experimental conditions and data measurements
– “Adjust” experimental measurements to as if they were
collected under the same conditions
– More experimental conditions (like recording solution
contents) can be extracted; improved metadata models in
future
• What new things can we learn about functional
relationships between different neuron types?
– How are biophysical properties correlated?
– Are there unknown similarities among neuron types?
33

34. Exploring correlations among
biophysical properties
34
τ = RC

35. Exploring correlations among
biophysical properties
35

36. 3636
Neuron
types
Neuronclusteringonbasisof
electrophysiology

37. 3737
Neuronclusteringonbasisof
electrophysiology

38. 38
Neuronclusteringonbasisof
electrophysiology

39. Analyzing neuronal
electrophysiological diversity
• What new things can we learn about
functional relationships between different
neuron types?
– Electrophysiological properties are highly
correlated across neuron types
– NeuroElectro provides a novel platform for
generating hypotheses on functional similarities of
neuron types
39

40. Outline
• Background on neuron electrophysiology
• The role of neuron variability
• NeuroElectro: a window to the world’s
neurophysiology data
• Analyzing the electrophysiological diversity of
neurons throughout the brain
• Extending NeuroElectro and future directions
– Validating the data contained within NeuroElectro
• Obtaining data at higher resolution
– What is the mechanistic basis of neuron diversity?
• Integration with Allen Institute Gene Expression Atlas
40

41. Validating and extending NeuroElectro
41
Shawn Burton, CMU Biology
CA1, pyr
Ctx, pyr (L5)
MOB, mit
Ctx, bskt

42. Validating and extending NeuroElectro
42
Shawn Burton, CMU Biology
CA1, pyr
Ctx, pyr (L5)
MOB, mit
Ctx, bskt

43. Validating and extending NeuroElectro
43Tripathy et al 2013
de Waard et al 2013 in press

44. Extending NeuroElectro and future
directions
• Data collected using Urban Lab specific
protocols is consistent with NeuroElectro data
– Currently exploring how to integrate collected raw
data into NeuroElectro
– NeuroElectro provides an easy check on data
quality
• What is the mechanistic basis of neuron
diversity?
– Integration with Allen Institute Gene Expression
Atlas
44

45. What is the mechanistic basis of
neuron electro-diversity?
45
Central dogma of
biology

46. Whole-genome correlation of gene
expression and electro-diversity
20,000 genes
46
Allen Gene Expression Atlas; Lein et al 2007
Systematic
variation among
neuron types
Patterns of gene
expresion
Electrophysiological
phenotypes

47. Mapping neuron electrophysiology to
gene expression
47
20,000 genes
Neuron type
resolution
Cell layer
resolution
Neuron type to cell layer mapping is
approximate. Will be improved in future
iterations with high resolution data.
Neocortex L5/6
pyramidal cell
Neocortex layer
5/6
Neocortex
basket cell
Neocortex

48. Correlation of neuronal
electrophysiology and gene expression
Electrophysiological differences
48
Pearson’s r = .36; p < 6*10-5
Neurontypes
Gene expression differences
(voltage gated ion channel genes)
Celllayers/
brainregions

49. Correlation of neuronal
electrophysiology and gene expression
49
Functional gene classes from Gene Ontology project
Ashburner et al 2000
not expressed
in brain
ion channel
gene classes

50. Extending NeuroElectro and future
directions
• Data collected using Urban Lab specific protocols is
consistent with NeuroElectro data
• What is the mechanistic basis of neuron diversity?
– Integrating NeuroElectro with the Allen Gene Expression
Atlas allows asking how gene expression influences neuron
biophysics
• Generates new hypotheses on link between specific gene classes
and neurophysiology
– Future gene expression datasets at higher cellular
resolution will allow testing of more specific hypotheses
– Fleshing this out is a large part of my proposed work as a
post-doc in Paul Pavlidis’s lab in Vancouver
50

51. Summary/Take home message
• Neurons are different
and differences
underlie diversity of
functions
51

52. CA1,pyrCtx,pyr(L5)
Why NeuroElectro?
• We need a “parts list” of
the brain
– BRAIN Initiative:
“consensus of cell types”
52

53. Why NeuroElectro?
• How much “buried
treasure” is in the
literature?
– What can we learn by
putting together what
we already know?
53
20,000 genes

54. Why NeuroElectro?
• The web lets us
“share data” and
ideas between
researchers in a way
that journal articles
do not
54

55. Acknowledgements
• People
– Nathan Urban
– Krishnan Padmanabhan
– Rick Gerkin
– Shawn Burton
– Urban Lab (past and present)
• Institutions
– Center for the Neural Basis of
Cognition
– Allen Institute for Brain Science
– Neuroscience Information
Framework
– Elsevier Research Data Services
– International Neuroinformatics
Coordinating Facility
– Open Source Software Community
• Funding
– NSF Graduate Research Fellowship
– RK Mellon Foundation
55

56. 56

57. Summary of NeuroElectro contents
80 neuron types; 300+ articles
57

58. Cross-validation for metadata
correction
58

59. Robustness of metadata correction
analysis
59

60. Projecting neurons onto low
dimensional space
60

61. Robustness of electrophysiological
neuron similarity analysis
61

62. Top 30 most biophysically correlated
gene classes
62

63. 63

64. Big science versus small hypothesis-
driven science

65. Mechanistic explanations for neuronal
biophysics
• The traditional
approach:
Iberiotoxin, a BK ion channel blocker
Extracted from the Indian Red Scorpion
65
Pharmacological drug

66. Defining neuron variability
Padmanabhan and Urban, Nat. Neuro. 2010
66
Krishnan Padmanabhan
Now at Salk Institute
Olfactory bulb mitral cells
(named because their shape looks like
a bishop’s hat, or mitre)

67. Defining neuron variability
1. Define a precise way of quantifying
neuron electrophysiology differences
2. What is the computational role of
electrophysiology differences?
67

68. The olfactory neural circuit
Volatile odor molecules
Nasal epithelium
Olfactory receptor
neurons
Olfactory Bulb Mitral Cells To cortex
~25-50 per glomerulus
Olfactory Bulb
68
Glomeruli

69. Neuron modeling approach
Neuron input
(current)
Neuron output
(action potentials)
A “model what you see”
based approach
69

70. Modeling mitral cell responses
70
Tripathy et al, in review

71. Modeling mitral cell responses
71

72. Models capture mitral cell biophysical
variability
72

73. Dimensionality reduction for
visualization
Dimensionality Reduction
(via PCA)
Neuron dimension 1Neurondimension2
73

74. Dimensionality reduction for
visualization
74

75. Dimensionality reduction for
visualization
1. Define a precise way of quantifying
neuron electrophysiology differences
2. What is the computational role of
electrophysiology differences?
75

76. Computational role of neuron
variability: Approach
76

77. Computational role of neuron
variability: Key results
77
Tripathy et al, in review

78. Importance of specific metadata
78

79. Central dogma of molecular biology
79

80. Lots of great tools for data sharing…

81. Barriers to data sharing
• Social
– “What’s in it for me? How will I get credit?”
– “It’s my data, not yours”
– “The benefit to me isn’t worth the time I put into it”
– “What if I get scooped?”
• Methodological
– “How do I share data? What do I share?”
– “Going back and annotating my files to share is super-
time consuming”
– Specifying file formats, data standards
– Building FTP servers and nice user interfaces

82. Project idea
• How can we make a standard neuroscience
wet lab more data-sharing savvy?
• Incorporate structured workflows into the
daily practice of a typical electrophysiology lab
(the Urban Lab at CMU)
– What does it take?
– Where are points of conflict?

83. Key insights/motivations
1. Effective data
sharing includes raw
data files +
experimental
metadata (typically
stored in a lab
notebook)
SDB_MC_12_voltages.mat

84. Key insights/motivations
1. Share raw data files
+ experimental
metadata
2. You know the most
about an
experiment when
you’re performing it

85. Key insights/motivations
1. Share raw data files +
experimental
metadata
2. You know the most
about an experiment
when you’re
performing it
3. Improved data
practices should
make labs more
productive

86. Project schematic

87. Metadata data app
• Electronic lab
notebook models
sequential slice-
electrophysiology
workflow
– Replaces pen-and-
paper lab notebook

88. Metadata data entry
• Electronic lab
notebook allows
structured data entry
Animal Strain

89. Metadata data entry
• Electronic lab
notebook allows
structured data entry
(i.e., dropdown
menus)
– Allows incorporation
of semantic ontologies
• Important to strike a
balance between
structure and
flexibility
MGI:3719486

90. Metadata data entry
MGI:3719486
• Electronic lab
notebook facilitates
entry of new content,
like registration of
recorded neurons to
brain atlas

91. Data integration
• Syncing of metadata
app and
electrophysiology data
acquisition via server
– Each trace of
experimental data
annotated with
metadata
• IGOR-Pro specific,
support pClamp, other
acquisition packages as
needed later

92. Data dashboard (web-based)

93. Data dashboard (future-steps)
• Use collected
metadata to sort
experiments
– Like mouse strain,
neuron type, animal
age
• Enable in-browser
analyses
– Track provenance
of analyzed data
back to raw data

94. Next steps
• Use built tools
– Populate data server with many experiments
• Is use of e-notebook too prohibitive?
– If yes, continue to iterate
– What can we ask now that we couldn’t before?
• It is much easier to ask exploratory questions, like
– How is the cell type that Shawn records different from the one that Matt
records?
• Exposing data to neuroscience databases
– NIF, INCF Dataspace, neuroelectro.org
• How adaptable are these solutions for use in other
labs?
• Who is going to pay for this?