Jeffrey C. Erlich

How is chaotic neural activity, driven by internal dynamics and external sensory input, resolved into coherent behavior?

Animals have many competing goals and drives as well as a barrage of sensory input to process. If our attention and actions are as frenetic as the world around us (as may be the case in attention deficit disorder) we will have difficulty accomplishing our goals. How does the brain deal with all of this competing input? How do brain structures deal with ambiguous or conflicting sensory information? And how do different brain structures communicate, influence and compete with each other so that the result of this competition is coherent thought and action? These questions cover a range of topics: attention, decision-making, cognitive control, planning, working memory, and others.

Erlich Lab Research Projects

Mapping decision-coding in the cortex of rats

The rat is a relatively understudied animal model for decision-making. But due to their size, cost and cognitive abilities they are a powerful model. During my postdoctoral work in Carlos Brody's lab I helped identify several key areas (e.g. the rat frontal orienting field and posterior parietal cortex) that show decision-related activity. However, the borders of these areas are unknown, and similar activity may be found elsewhere in the cortex. In my lab, we will further map out using electrophysiology and optogenetics the encoding of probability, value, effort, confidence and action across rodent cortex and examine the neural mechanism of competition or trade-off between factors (e.g. choosing a large effortful reward vs. a small easy reward).

Stress and decision-making

We all experience conflict between immediate rewards (a delicious pastry) and long-term goals (staying fit). One important variable in comparing immediate and future rewards is called the discounting factor: how much we undervalue future rewards. In humans, it has been shown that stress can increase the discounting factor, leading to more choices of immediate rewards. This phenomena can have serious social consequences. Those living in poverty experience significant stress that can perpetuate the cycle of poverty by driving decisions to be more steeply discounted (Haushofer & Fehr, 2014). One project in my lab will be the development of a rodent model of economic decision-making under stress to determine the neural mechanisms by which stress exacerbates discounting.

Reinforcement learning of deep networks (with Prof. Zheng Zhang)

The Deep Mind team (now at Google) has made recent progress in training neural networks with sparse reinforcement (as opposed to the more typical supervised learning algorithms). We are building on that work to develop an network model of a rat that will be trained on the same tasks that we use to train our real rat. This will be a test bed for our theories about learning, reinforcement, and neural mechanisms of cognition and will allow us to conduct experiments in silico, which we can then test in vivo.

Postoctoral Research Projects

Memory-Guided Orienting

Our memory-guided orienting task is modeled after the classic memory-guided saccade paradigm. Using this task, we demonstrated that the medial agranular cortex (AgM) a structure previously hypothesized, based on anatomy and lesion studies, to be a homologue of the primate frontal eye field (FEF), shows striking similarity to the FEF both in short-term spatially specific persistent activity and also in the effects of temporary inactivation (Erlich et al, 2011).

Pro/Anti Orienting

A classic task for the analysis of executive control is the anti-saccade task (Munoz, 2004). This task, until now, has been exclusively studied in primates (humans and monkeys). In the Brody lab we have trained rats on a pro/anti orienting task where we instruct the animal on each trial whether the rule is pro or anti and then a target appears to the left or right. On pro trials the animal orients towards the target for reward. On anti trials the animal orients away from the target for reward. I have collected data from frontal cortex (both medial prefrontal cortex and the frontal orienting field) and the superior colliculus from this task. I am in the process of collecting additional neural recordings from animals that are also implanted with infusion cannula.

Poisson Click Accumulation Task

Our auditory click integration task is loosely modeled on the random dot-motion task that has been used extensively to study decision-making in primates. The rat has to sit still and listen to a stream of clicks coming from the left and right speakers. In order to maximize reward the rat must count the clicks coming from the left and right speaker and decide whether there were more on the left of the right. Careful analysis and modeling of the behvior of the rats in this task has confirmed that rats indeed perform the task by counting or integrating the clicks (Bruton & Brody, submitted).

I have performed muscimol inactivations of the frontal cortex, AgM in particular, and the posterior parietal cortex (PPC) of rats performing this task. To our knowledge, these are the first frontal and parietal cortex inactivations in animals performing an accumulation of evidence task. Our results, consistent with lesion data, show that unilateral inactivation of frontal cortex result in profound contralateral impairment. That is, rats were significantly more likely to make a response ipsilateral to the infusion site. Surprisingly, inactivations of the parietal cortex had weak and inconsistent effects on behavior. It was only when the AgM was bilaterally inactivated that unilateral inactivation of the PPC resulted in a contralateral impairment. This suggests that for this integration of information task the PPC plays a secondary role to the AgM. This is consistent with recent work from primates showing that PPC inactivations have little effect on instructed choice (Wilke, Kagan & Andersen, 2012). This work is currently being prepared for resubmission.

Flash Counting Task

This task is similar in spirit to the Poisson Click task but in the visual domain. While the subject holds still a sequence of flashes appear on the left and right. A reward is available on the side with the greater number of flashes. Behavioral analyses of rats performing this task showed that subjects used a similar strategy on this visual accumulation task as in the Poisson Clicks task. In collaboration with David Tank's lab, rats have been performing this task during voluntary head fixation in order to record neural activity using 2-photon imaging.

Chaos Pendulums

Since 2009, I have taken up metalwork as a hobby. For my first project, I machined a pair of double pendulums, now refered to as the "classics", to bring to an art festival. In 2010 I upgraded the classics, learning from mistakes I had made the first time around. One aspect of the upgrade was to use a steel axle instead of an aluminum axle. This provided much more rigidity and so less energy was lost from wobbling. Another upgrade was to embed an LED and battery inside the outer pendulum. At night, persistence of vision provides a very nice trail of light, which can be approximated in photos using a high aperture and long exposure time.

After the success of the classics in 2010, I wanted a new challenge. Aluminum is a great metal in the machine shop. It is lightweight and relatively soft. However, with hand controlled mills and lathes anything shape other than a bar or cylinder was pretty tricky. So I signed up for a metalworking class at the School of Visual Arts (SVA) in New York to learn how to weld and shape steel. Unfortunately, there was no furnace at the studio at SVA so we were limited to cutting, with saws, and my favorite, the plasma cutter. A plasma cutter lets you cut through a sheet of steel like a hot knife through butter. Although the studio had a computer controlled plasma cutter, I did all my cutting by hand. Using a MIG welder and the plasma cutter I made the octostar double pendulum sculpture octostar pendulum as a gift to friends in Brooklyn.

Publications

Kopec, C.D., Erlich, J.C., Brunton, B.W., Deisseroth K, and Brody C.D. (2015) Cortical and Subcortical Contributions to Short-Term Memory for Orienting Movements. Neuron, doi:10.1016/j.neuron.2015.08.033 link

Duan C.A., Erlich J.C. and Brody C.D. (2015) Requirement of Prefrontal and Midbrain Regions for Rapid Executive Control of Behavior in the Rat. Neuron, doi:10.1016/j.neuron.2015.05.042 link

Erlich J.C., Brunton B.W., Duan C.A., Hanks T.D. and Brody C.D. (2015) Distinct behavioral effects of prefrontal and parietal cortex inactivations on an accumulation of evidence task in the rat. eLife, 4:e05457. link

Hanks, T.D., Kopec, C.D., Brunton, B.W., Duan, C.A., Erlich, J. C., and Brody, C. D. (2015) Differential roles in decision-making for accumulation-correlated neurons in parietal and prefrontal cortex. Nature, 520: 220-223. link

Erlich, J. C., and Brody, C. D. (2014). What to do and how. Nature, 503: 45-47. link

Erlich, J. C., Bush, D. E. A., and Ledoux, J. E. (2012). The role of the lateral amygdala in the retrieval and maintenance of fear-memories formed by repeated probabilistic reinforcement. Front Behav Neurosci, 6: 16. link

Erlich, J. C., Bialek, M., and Brody, C. D. (2011). A cortical substrate for memory-guided orienting in the rat. Neuron, 72: 330-343. link

Pai, S., Erlich, J. C., Kopec, C., and Brody, C. D. (2011). Minimal impairment in a rat model of duration discrimination following excitotoxic lesions of primary auditory and prefrontal cortices. Front Syst Neurosci, 5: 74. link

Ross, B. M., Moszczynska, A., Erlich, J., and Kish, S. J. (1998a). Low activity of key phospholipid catabolic and anabolic enzymes in human substantia nigra: possible implications for Parkinson's disease. Neuroscience, 83: 791-798. pdf

Ross, B. M., Moszczynska, A., Erlich, J., and Kish, S. J. (1998b). Phospholipid-metabolizing enzymes in Alzheimer's disease: increased lysophospholipid acyltransferase activity and decreased phospholipase A2 activity. J Neurochem, 70: 786-793. pdf

Ross, B. M., Hudson, C., Erlich, J., Warsh, J. J., and Kish, S. J. (1997). Increased phospholipid breakdown in schizophrenia. Evidence for the involvement of a calcium-independent phospholipase A2. Arch Gen Psychiatry, 54: 487-494. pdf

Other Writing

Jeffrey C. Erlich (2006) Ph.D. Thesis. To Fear or Not to Fear: The role of the amygdala & prefrontal cortex in the regulation of fear.pdf

Jeffrey C. Erlich (1997) The Recovered Memory Debate. pdf

Industrial Design

During my graduate work I became very frustrated with the cost and complication of moveable microdrives. When I arrived in Princeton, I researched a design for a cheap and easy to build design. I found a design based on a bic pen. I heavily modified the design, and after several iterations and feedback from lab members, we have a very light, very cheap moveable microdrive that has been used to successfully record from many thousands of neurons in the Brody lab, including individual animals with hundreds of single units over the course of months. The details of the design and the surgical techniques that result in months of viable recording are being written up as a methods paper.

Programming

MATLAB

MATLAB is the programming environment of choice in the Brody Lab (and really neuroscience). I have written thousands of lines of MATLAB code and it is certainly the programming language that I am currently most fluent in. MATLAB has obvious benefits. As an interpreted language where a matrix is a primitive data type, development for analysis of data is extremely fast. There are thousands of built in functions and features that allow the quick visualization of data. It is fairly easy to read and debug. However, there are lots of things not to like about MATLAB. First, it is expensive and proprietary. Many people in the scientific community prefer the FOSS R or Python with SciPy, NumPy and MatplotLib. I recently became more acquainted with R and knitR, specifically to use the lme4 package for generalized linear mixed models. I have had a little bit of exposure to Python as well, but I would not say that I am fluent in either language.

Shell scripts and Unix commands

BASH shell scripts using unix commands such as sed, awk, find, and grep are my goto for file manipulation and computer administration. My favorite thing about MacOS is that underneath all the gloss is a very Unix like OS which can be accessed through the terminal.

The Brody lab also has several linux servers that host our subversion code repositories, video archives, wiki, web server, and MySQL databases. Knowing my way around BASH shell has come in very handy.

Java

Java is a fantastic FOSS language with strict object-oriented rules. Non-graphical java code runs fairly seamlessly across platforms, and is a good choice for cross-platform development. My work at Nevo was all in Java, but since then my java skills have gotten a bit rusty. I recently refreshed my Java skills writing a cross-platform java client that runs on the linux computers in the behavioral training facility to help us monitor and maintain those computers. The monitoring system was written by another postdoc in the lab, Chuck Kopec, originally for windows machines only.

Other Languages

During my time as a computer science undergraduate I was exposed to many other programming languages: Pascal, C, Scheme, and 68040 Assembly. Notably, I wrote an artificial intelligence checkers program, Little Wing, in C. Little Wing was undefeatable since it would search about 14 moves ahead in a few seconds on a 1997 computer. It did end many games in a stalemate.

Relational Databases

During my year at Nevo I became aquainted with relational databases (Oracle, specifically). When I returned to academia, i began to use MySQL to store, organize and analyze my data. Additionally, I wrote a message passing system using MATLAB and MySQL that allowed me to run analysis jobs on any machine that could run MATLAB, essentially turning the NYU neuroscience network into a computing grid. This allowed me to run statistical analyses in hours instead of weeks.

When I joined the Brody lab, MySQL was an obvious choice for helping to administer the high-throughput behavioral training facility. I also refined the message passing system into the Brody Computing Grid, so that the computers that are used for training rats during the day can be used for analysis and simulation at night. Best of all MySQL is free and open source (FOSS) software. It was acquired by Oracle in 2009 which sparked the fork of MySQL into MariaDB which is a binary equivalent of MySQL but is completely GPL and not affiliated with Oracle.

I'm very interested in non-relational databases such as SciDB which are designed for large continuous data sets, such as astronomical data. This kind of database would be appropriate for storage and analysis of continuous electrophysiological data or video. Unfortunately, I have not yet had the time to install it and try it out. If you have experience with SciDB and neuroscience get in touch!.

Multiple positions are available in my lab at NYU-Shanghai.

To apply for any position include a letter explaining your interest in the lab and in the position, your C.V. and 2-4 letters of reference to apply at erlichlab dot org

Postdoctoral Fellow

Multiple postdoctoral positions are available starting April 1, 2014 in Jeffrey Erlich's lab at NYU-Shanghai as part of the new NYU-ECNU Institute of Brain and Cognitive Science. We study the neural basis of cognition using sophisticated behavioral paradigms combined with electrophysiology, pharmacology, optogenetic and computational/theorectical methods. As founding members of the lab, you will contribute to the setup of a high-throughput training facility for rodents, high-channel count awake behaving neurophysiology, and high throughput optogenetic screening of neural circuits involved in cognitive function. English fluency required. PhD in neuroscience or related field required. Experience with rodent behavioral neurophysiology prefered.

Graduate Students

I have room in my lab for a few enthusiastic graduate students. There are two ways to join my lab as a graduate student. Either via the NYU CNS Shanghai Track or through East China Normal University. Interested applicants should send a C.V. and a cover letter addressing their interest in the lab to apply@erlichlab.org

Principal Investigator

Postdoctoral Fellows

Staff

Students

Collaborators