Command-line Electrophysiology

A Closed-loop Approach to Single Cell Characterisation

Jo˜ao Couto

, Daniele Linaro

and Michele Giugliano

1,2,3

Theoretical Neurobiology and Neuroengineering, Dept. Biomedical Sciences, Univ. of Antwerp, B-2610 Wilrijk, Belgium

Dept. Computer Science, Univ. of Shefﬁeld, S1 4DP Shefﬁeld, U.K.

Laboratory of Neural Microcircuitry, Brain Mind Institute, EPFL, CH-1015 Lausanne, Switzerland

1 OBJECTIVES

Neuronal electrophysiology experiments are typically

performed by using expensive commercial software

packages and/or custom written applications: the lat-

ter are usually suited only for particular experiments

and lack documentation. Additionally, the currently

available packages, while focussing on usability by a

large community of scientists, fail to provide proper

scripting methods to systematically span the parame-

ter space or to apply standard optimisation techniques

while performing the experiment.

As neuroscience progresses, scientists are faced

with questions that can only be answered by precisely

triggering the stimulation to ongoing features of the

experiment. This implies the usage of experimental

paradigms that rely on short latencies (Prinz et al.,

2004; Wallach et al., 2011) and whose implementa-

tion may require particular hardware and/or can be

time consuming to develop from scratch. Recently,

there has been growing interest in novel experimen-

tal paradigms based on real-time systems to perform

closed-loop stimulation (Bettencourt et al., 2008; Lin

et al., 2010; Zrenner et al., 2010): however, these ap-

proaches lack modularity and ease of integration with

general scripting languages as they heavily rely on

graphical user interfaces.

We developed a ﬂexible software platform called

LCG, whose main objective is to enable scientists to

abstract from the technical details of the experiment

while providing simple yet powerful tools to analyse

online the recorded data and react accordingly.

LCG was developed with the following goals in

mind:

• Serve as a general abstract platform for single cell

and network electrophysiological experiments.

• Implement dynamic clamp (see (Robinson and

Kawai, 1993; Sharp et al., 1993; Prinz et al.,

2004) for a review) with active electrode compen-

sation (Brette et al., 2008) to allow single elec-

trode experiments.

• Allow relatively easy implementation of closed

loop and hybrid experiments.

• Support multiple real-time engines in a standard

desktop computer.

• Provide a simple installation procedure and sup-

port for scripting for on-the-ﬂy parameter optimi-

sation.

In the following sections we present the main features

of LCG, which are relevant for performing in vitro or

in vivo experiments employing both intracellular and

extracellular recordings. As an application example,

we describe a way to efﬁciently characterise the input-

output relationship of single cells using a closed loop

approach.

2 METHODS

The toolbox is written in C/C++ and consists of a li-

brary that can be linked to C/C++ programs written

by the user. The software uses COMEDI to interface

with the data acquisition card and supports two real-

time Linux kernels, PREEMPT-RT and RTAI. It is

worthwhile noting that, for performing standard volt-

age or current clamp experiments, the system does not

require a real-time kernel.

The compilation of LCG generates two command-

line applications that allow the user to perform non

real-time (i.e., classical voltage and current clamp and

extracellular recordings) and real-time experiments,

respectively. This also implies that users do not need

to have any knowledge of a programming language

to effectively use LCG. The choice of a command-

line interface (CLI) over a graphical user interface of-

fers speed (an experienced user can perform the same

task much faster using a CLI), stability (CLIs do not

change between software updates) and scripting ca-

pabilities, enabling the user to automate a sequence

Couto J., Linaro D. and Giugliano ..

Command-line Electrophysiology - A Closed-loop Approach to Single Cell Characterisation.

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

of tasks by including them in a script. Additionally,

the real-time program uses XML conﬁguration ﬁles

that (i) allow to programmatically explore physiolog-

ical parameter ranges from a script, (ii) assure repro-

ducibility since the conﬁguration ﬁle can be stored

with the recorded data and (iii) allow the development

of complex protocols.

2.1 Toolbox Design Considerations

LCG is based on the concept of interacting entities:

these are objects that can have multiple inputs, have

one single output and can be connected to each other.

Importantly, they implement the necessary logic to

produce the output at each time step of the experi-

ment. By making that every class in LCG is an entity

and assuring that each entity performs only one spe-

ciﬁc task, code reusability is maximised, since com-

plex experiments can be designed by combining mul-

tiple entities together. Entities can also communicate

with each other asynchronously by emitting events:

these are delivered to all connected entities and can be

used, for instance, to communicate the occurrence of

a spike or to send trigger and reset messages to other

entities.

While it is possible to implement new custom entities

in C++, LCG provides a set of entities that allow to

perform common tasks such as data input and output,

storing recorded data in compressed format, and gen-

erating arbitrary (current/voltage/conductance) wave-

forms. To perform hybrid experiments, LCG comes

with a set of entities that implement commonly used

neuronal and synaptic models.

3 RESULTS

We now discuss a simple application where LCG was

used to record from pyramidal neurons in acute brain

slices of the somatosensory cortex of juvenile rats.

One of the ways to characterise the input-output

relation of single neurons is by computing the current-

frequency relationship (f-I curve). This is tradition-

ally done by applying several DC current steps of var-

ious amplitudes and measuring the corresponding ﬁr-

ing frequency. As one has to let the cell recover be-

tween trials it can take several minutes to compute an

f-I curve, and the curve is sampled coarsely since the

current injected in the neuron is constant during each

trial. We propose a novel approach to measuring the

input-output relation by performing the experiment

in closed loop using a proportional-integral-derivative

(PID) controller. A sketch of the experimental setup is

shown in Fig. 1A: a Real Neuron entity records the in-

200

400

600

f (Hz)

I (pA)

0.5nA

15Hz

0.2nA

5Hz

target

PID

Frequency

Estimator

Real Neuron

PID

Controller

Ramp

Figure 1: Example of computation of an f-I curve with a

PID controller. A Schematic representation of the experi-

mental setup (see text for explanation). B Example of the

frequency estimate (red) and of the injected current (blue)

to have the ﬁring frequency follow a frequency ramp rang-

ing from 5 to 30Hz (magenta trace). The blue curve is the

current injected by the controller. C Instantaneous ﬁring

frequency (inverse of the ISI) as a function of the injected

current (gray circles). The red line is a linear ﬁt to the data

and the green dots are the f-I curve computed using the stan-

dard method.

tracellular membrane potential of the cell under anal-

ysis and detects its spikes, which are sent, as events, to

the Frequency Estimator entity. The output of the es-

timator is sent to the PID controller, which compares

it to a target value, in this case a Ramp waveform.

The controller, in turn, computes the current required

to clamp the neuron to the target ﬁring frequency and

feeds it back to the neuron. The time course of the

estimated ﬁring frequency and of the injected current

are shown in Fig. 1B, while the computed f-I curve

is shown in Fig. 1C. The red curve is a ﬁrst degree

polynomial ﬁt to the inverse of the ISIs (gray circles)

and the green dots are the f-I curve computed with the

“traditional” method. The computation of an f-I curve

with the method presented here takes around 30 sec

and provides a much better sampling than the tradi-

tional method. This is possible because the output

of the PID is constant between spikes, thus allowing

to associate each inter-spike interval (ISI) to a spe-

ciﬁc (constant) value of current, building the f-I curve

shown here.

This closed-loop approach is substantially differ-

ent from the injection of a ramp of current to compute

the f-I curve: in fact, in the latter case, the current

injected between spikes is not constant, which makes

associating an ISI with a particular current value dif-

ﬁcult.

Additionally, this same experimental setup allows

to clamp the ﬁring frequency of a neuron at a given

value, which can be useful in protocols that require

delivering perturbations at ﬁxed phases in the ﬁring

cycle of the cell (Miranda-Dom´ınguez et al., 2010).

4 DISCUSSION

In this contribution we have presented LCG, a soft-

ware to perform open- and closed-loop electrophysi-

ological experiments. We have described some of the

general principles that underly code reusability and

ﬂexibility of the toolbox. As a test case, we have

shown a novel efﬁcient way of measuring the input-

output relationship of neurons by continuously esti-

mating the ﬁring frequency of the cell and, in closed-

loop, clamping it to a desired frequency value. This

allows to sample the f-I curve both with very high res-

olution and in a relatively short time (less than 30s).

We are currently using LCG to perform both

dynamic clamp experiments where very low laten-

cies are required and in vivo experiments combining

whole cell with extracellular recordings where large

populations of neurons are probed simultaneously.

Using LCG in conjunction with high-level scripting

languages such as Python or Matlab offers several ad-

vantages, among which we mention:

• the possibility to use standard optimisation proce-

dures to ﬁnd, in real-time, optimal parameter val-

ues.

• The possibility to implement an experimental

pipeline with standardised protocols, which allow

to easily compare cells across experimental condi-

tions and to speed up the successive data analysis.

• A relatively easy implementation of hybrid mi-

crocircuits, along the lines of what has been pre-

sented in (Kispersky et al., 2011).

It is worthwhile noting that, while some commercial

and open source packages offer scripting capabili-

ties, the possibility of performing electrophysiologi-

cal recordings at the command line opens new pos-

sibilities for automating experimental workﬂows and

allows users to closely integrate standard optimisation

tools (for instance from Python’s Numpy module) in

their experiments.

By enabling closed loop experiments at various lev-

els of latency and allowing to interface with general-

purpose scripting languages, LCG has the potential to

boost electrophysiologicalresearch to another level of

automation and protocol complexity with minimal ef-

fort on the neuroscientist part.

ACKNOWLEDGEMENTS

D.L. is supported by the Flanders Research Foun-

dation (grant no. 12C9112N, http://www.fwo.be).

This work was partly supported by the University of

Antwerp and by the European Commission, through

the Seventh Framework Programme under the ICT -

Future and Emerging Technologies scheme (project

ENLIGHTENMENT, grant agreement no. 284801).

REFERENCES

Bettencourt, J., Lillis, K., Stupin, L., and White, J. (2008).

Effects of imperfect dynamic clamp: computational

and experimental results. Journal of Neuroscience

Methods, 169:282–289.

Brette, R., Piwkowska, Z., Monier, C., Rudolph-Lilith, M.,

Fournier, J., Levy, M., Fr´egnac, Y., Bal, T., and Des-

texhe, A. (2008). High-resolution intracellular record-

ings using a real-time computational model of the

electrode. Neuron, 59(3):379–391.

Kispersky, T. J., Economo, M. N., Randeria, P., and White,

J. A. (2011). GenNet: A platform for hybrid network

experiments. Frontiers in Neuroinformatics, 5:11.

Lin, R., Bettencourt, J., White, J., Christini, D., and Butera,

R. (2010). Real-time experiment interface for biolog-

ical control applications. In Engineering in Medicine

and Biology Society (EMBC), pages 4160–4163.

Miranda-Dom´ınguez, O., Gonia, J., and Netoff, T. I.

(2010). Firing rate control of a neuron using a lin-

ear proportional-integral controller. Journal of Neural

Engineering, 7(6):066004.

Prinz, A. A., Abbott, L. F., and Marder, E. (2004). The dy-

namic clamp comes of age. Trends in neurosciences,

27(4):218–224.

Robinson, H. P. and Kawai, N. (1993). Injection of digitally

synthesized synaptic conductance transients to mea-

sure the integrative properties of neurons. Journal of

Neuroscience Methods, 49(3):157–165.

Sharp, A. A., O’Neil, M. B., Abbott, L. F., and Marder,

E. (1993). The dynamic clamp: artiﬁcial conduc-

tances in biological neurons. Trends in neurosciences,

16(10):389–394.

Wallach, A., Eytan, D., Gal, A., Zrenner, C., and Marom, S.

(2011). Neuronal response clamp. Frontiers in neuro-

engineering, 4:3.

Zrenner, C., Eytan, D., Wallach, A., Thier, P., and

Marom, S. (2010). A generic framework for real-time

multi-channel neuronal signal analysis, telemetry con-

trol, and sub-millisecond latency feedback generation.

Frontiers in neuroscience, 4:173.