>> WELCOME TO NIH AND THE RESEARCH SYMPOSIUM. I'M JIM ANDERSON AND DIRECT THE DIVISION IN THE OFFICE OF THE DIRECTOR, INCLUDING THE COMMON FUND ABOUT OTHER ACTIVITIES AND I MET SOME OF YOU. LOOK FORWARD TO MEETING MORE AND HEARING YOUR GREAT SCIENCE. I WANT TO REMIND YOU THAT THE HR-HR PROGRAM GUIDE IS HERE, WHOM YOU ALL KNOW AND I WORK CLOSELY WITH HIM. THIS IS REALLY A ONE-OF-A-KIND MEETING FOR SEVERAL REASONS. ONE, BECAUSE OF THE DIVERSITY OF THE SCIENCE. YOU'RE NOT GOING TO SEE THIS BREATH OF THE SCIENCE IN ANY SINGLE MEETING TYPICALLY. THE COMMON FUND IS CROSSCUTTING. IT DOESN'T ACKNOWLEDGE IC BOUNDARIES. SO YOU'RE GOING TO MEET PEOPLE HERE WHO WOULD HAVE SPENT THEIR LIVES FUNDED BY ONE OF OUR INSTITUTES AND I MAY NEVER MEET THEM. SO THIS IS A GREAT OPPORTUNITY FOR NETWORKING. YOU WILL MAKE COLLEAGUES AND FRIENDS YOU WOULDN'T HAVE MET AND HOPEFULLY THEY CAN ENRICH YOUR CAREERS. SO, MY LAST JOB IS TO INTRODUCE DR. LAWRENCE TABAK, THE PRINCIPLE DEPUTY DIRECTOR OF NIH AND HE IS GOING TO PROVIDE A WELCOME AND ALSO INTRODUCTORY REMARKS. I'LL JUST SAY THAT LARRY IS A HEARTBEAT AWAY FROM THE DIRECTOR OF NIH. BUT ALSO HE'S A HEARTBEAT AWAY FROM THE COMMON FUND. HE WAS VERY INVOLVED BACK WHEN THE ROADMAP WHICH LED TO THE COMMON FUND OVER A DECADE AGO, AND HE IS STILL VERY INVOLVED IN BIG DECISIONS ABOUT DIRECTIONS WITH THE COMMON FUNDS. VERY INFORMED. LAYERY IS AN ACTIVE INVESTIGATOR TOO, SO, LARRY, THANK YOU FOR BEING HERE TODAY. [ APPLAUSE ] >> GOOD MORNING. SO, NORMALLY WHEN I SHOW UP AT A MEETING AT NIH, PEOPLE BECOME VERY DISAH POINTED BECAUSE IT MEANS THAT FRANCIS ISN'T GOING TO BE HERE. AND NO MATTER HOW HIGH I STAND ON MY TIPPY TOES, NO ONE WILL EVER MISTAKE ME FOR FRANCIS, RIGHT? SO BUT HE WILL BE WITH YOU LATER ON IN THE MEETING. AND UNFORTUNATELY, YOU'RE COMPETING WITH OUR SPECIAL CULTURAL LECTURE THIS MORNING, WHO HAPPENS TO BE YO-YO MA, SO, FRANCIS IS WITH YO-YO BUT WILL BE WITH YOU LATER ON. THE OTHER THING I'LL JUST MENTION VERY BRIEFLY, IS EVEN THOUGH THIS GROUP IS THE HIGH RISK, HIGH REWARD CADRE, NONE OF YOU IN THE BACK ARE WILLING TO TAKE THE RISK OF MOVING TO THE FRONT. [ LAUGHS ] SO COULD YOU ALL MOVE DOWN, PLEASE? GO AHEAD. PLEASE. DO IT. IT'S OKAY. COME DOWN. I PROMISE WE WON'T BITE. IF YOU DON'T MOVE DOWN, I WILL GIVE A LECTURE ON THE CREB CYCLE. [ LAUGHS ] THAT GOT THEM GOING. THAT'S GOOD. [ LAUGHS ] COME ON! IT'S OKAY. COME OUT OF THE DARKNESS INTO THE LIGHT. THIS IS MUCH BETTER. ROBBIE, ALL THE PEOPLE IN THE LAST THREE ROWS IF YOU WOULD CUT OFF THEIR FUNDING, PLEASE. [ LAUGHS ] THAT GOT A LITTLE MORE MOVEMENT. EXCELLENT. I GUESS THE PEOPLE WAY IN THE BACK THERE, ARE JUST HERE FOR SOME OTHER REASON? I DON'T KNOW. OKAY, SO THANK YOU ALL FOR BEING HERE. JIM ALREADY SAID THIS, BUT I DO WANT TO REEMPHASIZE THE POINT THAT THE COMMON FUND IS DIFFERENT. AND WE CAN'T REEMPHASIZE THIS ENOUGH, PARTICULARLY HERE AT NIH, WHERE EVERYTHING IS DRIVEN LARGELY BY INSTITUTES AND CENTERS. AND MANY OF YOU HAVE YOUR HOME INSTITUTE OR CENTER. THE COMMON FUND IS DIFFERENT BECAUSE IT TENDS TO BE A STIMULATOR. IT TENDS TO BE A CATALYST. IT TENDS TO DO THE STUFF THAT THE INSTITUTES AND CENTERS AREN'T EITHER QUITE YET WILLING TO TRY, OR BECAUSE IT'S THAT PERSON'S RESPONSIBILITY, THEY DON'T WANT TO TAKE IT ON. AND SO THE CHARACTERISTICS ARE DISPLAYED HERE, AND VIRTUALLY EVERY AWARD THAT EACH OF YOU NOW HOLDS, FITS MANY, IF NOT EACH OF THOSE, ATTRIBUTES. SO THAT IS REALLY A WONDERFUL THING. SO, NORMALLY WHAT I WOULD DO, IS I WOULD INTRODUCE EVERY SINGLE PERSON IN THE ROOM AND THAT WOULD TAKE ABOUT AN HOUR AND 50 MINUTES. SO I'M GOING TO DISPENSE WITH THAT. BUT SIMPLY, JUST QUICKLY RUN THROUGH THE DIFFERENT CATEGORIES OF AWARD, AND THEN ON BLOCK, FOR THOSE ON STUDY SECTION, YOU KNOW WHAT THAT MEANS; THEN ON BLOCK, WE'LL HAVE FOLKS STAND AND BE ACKNOWLEDGED. OKAY? SO THE FIRST IS THE PIONEER AWARD. AND THIS REALLY WAS THE THING THAT STARTED OFF THIS PORTFOLIO IN THE COMMON FUND, WHEN IT USED TO BE CALLED THE ROADMAP BACK, WAY BACK WHEN. IT BEGAN IN 2004 AND SO FAR, WE HAD 168 AWARDS. NOW, THAT SEEMS LIKE A LOT BUT WHEN YOU THINK OF THE THOUSANDS OF AWARDS WE RELEASE EACH YEAR, IT'S REALLY A VERY TINY, TINY NUMBER. 12 AWARDS THIS YEAR AND I BELIEVE WE HAVE SOME OF THE PIONEER AWARDEES HERE TODAY, SO IF YOU COULD PLEASE STAND TO BE RECOGNIZED. GREAT. THANK YOU. [ APPLAUSE ] THE NEW INNOVATOR AWARDS ARE THE ONES THAT START THOSE WHO ARE EARLIER IN THEIR CAREERS, AND WE BEGAN THIS IN 2007. THUS FAR, WE HAVE HAD 447 AWARDS WITH 48 NEW ONES FOR THIS YEAR. AND AGAIN, I'M GOING TO FLASH THROUGH THE RECIPIENTS, AND IF YOU COULD ALL STAND TO BE ACKNOWLEDGED, THAT WOULD BE GREAT. STOW, PLEASE SO PLEASE STAND IF YOU'RE A NEW INNOVATOR AWARDEE. NOT NECESSARILY IN THIS PARTICULAR SLIDE. [ APPLAUSE ] THANK YOU. THEN THERE IS THE TRANSFORMATIVE RESEARCH AWARD. THIS BEGAN IN 2009. THIS IS FOR THE SORT OF OUT-OF-THE-BOX, UNCONVENTIONAL, NO ONE WOULD BELIEVE IT IDEA, BUT IT'S RIGHT, IT IS GROUNDBREAKING. IT EXPLODES A FUNDAMENTAL PARADIGM. AND SO, HERE ARE THE TRANSFORMATIVE RESEARCH AWARDEES. IF YOU COULD STAND, PLEASE. [ APPLAUSE ] AND THEN FINALLY, AND I HAVE TO ADMIT THIS IS MY FAVORITE AWARD OUT OF ALL OF THEM. AND THE REASON IS, BECAUSE I NEVER HAD A POSTDOC. BECAUSE I DIDN'T KNOW YOU NEEDED TO HAVE A POSTDOC. THAT'S HOW DUMB I WAS BACK THEN. AND SO, BUT THAT WAS LIKE 100 YEARS AGO. BUT, TODAY OF COURSE, THE ROUTINE IS PEOPLE DO A POSTDOC. THIS BEGAN IN 2010. WE CALL IT THE SKIP THE POSTDOC AWARD. NOT EXACTLY; BUT IT ALLOWS PEOPLE TO BEGIN INDEPENDENT POSITIONS VERY EARLY IN THEIR CAREER. AND THIS IS NOT FOR THE FAINTHEARTED. YOU THINK BACK TO WHAT YOU LEARNED DURING YOUR POSTDOC, DURING THAT SORT OF SAFE HAVEN, I THINK YOU'LL AGREE THAT THIS AWARD IS NOT FOR THE FAINT OF HEART. SO FAR WE HAVE MID-'88 AWARDS AND WE HAVE 16 NEW ONES THIS YEAR. SO IF YOU COULD STAND TO BE ACKNOWLEDGED. [ APPLAUSE ] WITH THAT, UNDER TIME AND UNDER BUDGET, I WILL SEND IT BACK TO ROBBIE WHO LIN BEGIN THE PROGRAM. I HOPE YOU ENJOY THE SYMPOSIUM. [ APPLAUSE ] >> THANK YOU LARRY AND JIM FOR THOSE REMARKS AND I'D LIKE TO WELCOME YOU TO THE MEETING AS WELL. THE SCIENTIFIC EXTRAVAGAZA STARTS OUR FIRST SESSION, DR. SUE COASTER, THE 20 DIVISION DIRECTOR AT THE NATIONAL INSTITUTE OF MENTAL HEALTH WHO WILL MODERATE THE FIRST SESSION. SHE IS ALSO A MEMBER OF THE HIGH RISK, HIGH REWARD WORKING GROUP. >> GOOD MORNING, EVERYBODY AND I WANT TO ADD MY WELCOME TO THE NIH TO ALL OF YOU. AND ALSO TAKE A MINUTE TO THANK DR. TABAK FOR HIS SUPPORT OF THE PROGRAM AND HIS SERVICE TO THE NIH, WHICH AS THEY SAID, HE'S A HEARTBEAT AWAY FROM THE DIRECTORSHIP. WE WILL START OFF THIS MORNING WITH A SET OF TALKS ABOUT MY FAVORITE PART OF THE HUMAN BODY, WHICH IS THE BRAIN. OUR FIRST UP IS DR. ANDREAS TOLIAS FROM BAYLOR COLLEGE OF MEDICINE TELLING US ABOUT THE FABRIC OF NEOCORTEX CANONICAL STRUCTURE AND COMPUTATIONS. >> THANK YOU, EVERYBODY FOR COMING HERE TODAY AND THANK YOU FOR THE INVITATION. SO, THE NEOCORTEX IS THE PART OF OUR BRAIN INVOLVED IN HIGH-FUNCTIONS LIKE PERCEPTION, COMPREHENSION AND MOTOR CONTROL. AND IN THE LAST 100 YEARS, WE LEARNED A TREMENDOUS AMOUNT OF INFORMATION AT THE VERY SMALL SCALE, HOW INDIVIDUALS OPERATE AT THE MOLECULAR LEVEL AND ALSO AT THE MACRO SCALE, HOW THE GLOBAL ORGANIZATION OF THE BRAIN IS. HOWEVER, WE STILL DO NOT KNOW HOW BILLIONS OF NEURONS ORCHESTRATE THEIR ACTIVITY TOW PROCESS INFORMATION AND INTERACT IN A COMPUTATIONAL SENSE. AND THE HYPOTHESIS THAT I AND MANY OTHERS ASPIRE TO IS THAT THE NEOCORTEX IS COMPRISED OF REPEATING MODULES OR CORTICAL MOTIFS THAT CONTAIN NUMEROUS CELL TYPES ORGANIZED AND THEY IMPLEMENT CANONICAL OPERATIONS OR ALGORITHMS AND ULTIMATELY DISCOVERED THESE. SO, THE DEBATE IN THE FIELD WHERE THIS HYPOTHESIS IS RIGHT OR WRONG BUT I'LL GIVE YOU A MORE RECENT EVIDENCE MORE OF A PROOF-OF-CONCEPT THAT IT IS POSSIBLE TO HAVE A CANONICAL OPERATION ND DO PRETTY REMARKABLE THINGS. SO RECENTLY IN THE LAST TWO OR THREE YEARS, IT'S BEEN A REBS SANS IN THE FIELD OF AI WHY THE NEURONETWORKS ARE IMPACTING OUR EVERY DAY LIVES AND THESE DICES IN THIS ONE CASE WHERE THE NETWORKS COULD LABEL SA MANTICALLY. SO AT THE HEART OF THE PROBLEM OF VISION FOR EXAMPLE, IS THAT IF YOU LOOK AT THE IMAGES OF THESE TWO PHASES, THERE IS AN INFINITE NUMBER OF POSSIBILITIES. AND IN THE INPUT SPACE IT'S A HIGH-DIMENSIONAL MIND. WHAT THE BRAIN DOES IS UNFOLD THIS AND ULTIMATELY IN THE HIGH AREAS AND TEMPORAL LOBES, HAVE SOME OBJECT SELECTIVE PROPERTIES OR PHASES. WE WANT TO ADDRESS IN ORDER TO UNDERSTAND HOW THE BRAIN ACTUALLY PERFORM THIS OPERATION AND TO DECIPHER THE PRINCIPLES OF ORGANIZATION IN THIS CORTICAL MOTIFS. THE FIRST ONE IS WE WANT TO IDENTIFY ALL THE CELL TYPES THAT COMPOSE THE CORTICAL MICROSURGERY. BELIEVE IT OR NOT, UNLIKE OTHER PARTS OF THE BODY, WE STILL DON'T KNOW THE CELL TYPES THAT MAKE UP THE CORTEX. WE WANT TO KNOW THE WIRING AND WE WANT TO KNOW WHAT DO THESE CORTICAL MICROCIRCUITS COMPUTE? FOR THE CELL TYPE, THE APPROACH WE HAVE BEEN FOLLOWING IS THAT WE ADAPTED A MULTI-PATCHING METHOD WHERE WE PATCH IN SLICES MULTIPLE NEURONS AND THEN A POSTDOC IN THE LAB COULD DEVELOP THIS METHOD TO DO A FINE RECONSTRUCTION FILLING IN OF THE FINE DETAIL OF THESE NEURONS. AND RECENTLY WE IDENTIFIED, STUDIED LAYERS OF THE PRIMARY VISUAL CORTEX OF THE MOUSE AND IDENTIFIED CELL TYPES BASED ON MORPHOLOGICAL CRITERIA. NOW THESE CELL TYPES, SOME OF THEM WERE DESCRIBED FOR THE FIRST TIME IN THIS PAPER WE PUBLISHED AND THEY HAVE VERDICT MORPHOLOGY THE WAY THEY PROJECT BUT WE DON'T KNOW THEIR FUNCTION. AND LIKE MORPHOLOGY AND ELECTROPHYSIOLOGY, CANNOT DISTINGUISH THESE CELL TYPES. SO IN THIS CASE, PROVIDES US THE GOLD STANDARD. TRANSGENIC MICE WERE STUDIED AND WHAT WE FOUND IS THAT THE CURRENT AND THIS IS TO BE EXPECTED, CURRENT STANDARD OF MOLECULAR CLASSES THAT PEOPLE USE ARE PROMISCUOUS IN CELL TYPES, FOR EXAMPLE IN THIS CASE SOME OF THE STARTING LINES, WOULD LABEL BOTH. SO WHAT WE DECIDED TO DO, WE NEEDED A METHOD TO DO A MORE PRECISE QUANTIFICATION OF DIFFERENT CELL TYPES THAT COMPRISE THE CIRCUITS. AND TO DO THIS, WE DEVELOPED A STUDENT IN MY LAB, AT THE CAROLINS CAINSTITUTE, WE DEVELOPED THIS METHOD CALLED PATCH SEQ WHERE AFTER WE PATCH CELLS, WE WILL DO WHOLE-CELL TRANSCRIPT OMICS. SO FROM THESE, WE HAVE THE FIRING PATTERNS OF THE NEURONS, MORPHOLOGY AND WE CAN NOW RELAY THESE RESPECTS AND TRY AND NOT JUST FIND UNIQUE MARKERS FOR MORPHOLOGICALLY-DEFINED CLASS SYSTEM BUT GET A COMPLETE FEATURE OF HOW THE CIRCUIT IS ORGANIZED. TO CUT A LONG STORY SHORT, THIS IS PUBLISHED AND WE HAVE SHOWN IN LAYER ONE WHERE WE DEVELOPED THIS METHOD, THE TWO MAJOR CLASSES OF NEURONS WHEN YOU LOOK AT THE TRANSCRIPTOMICS SPACE AND YOU USE A CLUSTERING IN THAT GENE SPACE, YOU GET TWO DISTINCT CLASSES WHICH MAP VERY NICELY TO THE MORPHOLOGICAL CLASSES. THIS IS VERY EXCITING BECAUSE NOW WE HAVE THE FULL TRANSCRIPTOME WITH THIS METHOD WE GET UP TO 7000 GENES PER CELL AND NOW WE CAN LOOK AT WHAT IS DIFFERENT CELL, WHAT DIFFERENT TRANSCRIPTOMICS MEANS. JUST TO GIVE YOU SOME INTERESTING DIRECTIONS HERE FOR EXAMPLE, IF YOU DO ENRICHMENT, YOU SEE WHICH PATHWAYS ARE ENRICHING ONE VERSUS THE OTHER GENE AND FIND THESE CELLS IS MORE INVOLVED IN THINGS THAT HAVE TO DO WITH MITOCHONDRIAL FUNCTION WHICH AS THE SINGLE CELL IS MORE HAS TO DO WITH TRANSMEMBRANE PROTEINS AND ION CHANNELS. AND IF YOU LOOK AT THESE TWO CELL TYPES, YOU GET A INDICATION BASED ON THE ROLE THEY PLAY. LACKING PARTICULAR EXTRACELLULAR PROTEINS TO GIVE SPECIFICITY DURING DEVELOPMENT. WE HAVE FOUND GENES DIFFERENTIALLY EXPRESSING ONE CELL TYPE AND NOT THE OTHER. SO NOW WE CAN LINK THE GENETIC LEVEL OF LINKING HUMAN DISEASES TO GENES TO PARTICULAR CELL TYPE PATHOLOGY WHICH WOULD BE INTERESTING IN OPENING OTHER NEWS FOR PHARMACEUTICAL INTERVENTIONS OF THIS CELL TYPE SPECIFICITY. SO, THIS IS OUR APPROACH TO DEFINE THE CELL TYPES. WE USE MORPHOLOGICAL AND TRANSCRIPTOMIC AND NOW WE WANT TO DEVELOP AND SENSE ALL THE CELL TYPES THAT MAKE UP THE NEOCORTEX. IF YOU WANT AT THE TABLE. BUT THE NEXT STEP IS TO TRY TO UNDERSTAND HOW THEY ARE WIRED TOGETHER. HOW THESE DIFFERENT CELL TYPES WIRE TOGETHER AND THE HYPOTHESIS IS IT CANONICAL MICROCIRCUIT MOTIF. AND THERE IS MANY DIFFERENT APPROACHES TO DO THIS ANALYSIS AND WE ARE DOING MULTI-PATCHING. WE STIMULATE ONE AT A TIME AND WE GET 8 BY 8 OR 12 BY 12 AND WE COLLAPSE MANY OF THEM. AND WE COLLECT 11,000 PAIRS OF NEURONS THAT SHOULD DEFINE BASED ON THE CELL TYPE MORPHOLOGY AND NOW COMPUTED WIRING DIAGRAM OF ALL THE 16 CELL TYPES PLUS THE TWO PARAMEATAL CELLS IN LAYER 5 AND THAT WIRING LOOKS LIKE. THIS SO IF YOU LOOK AT THIS, IT LOOKS IF I JUST STOP THE SLIDE AND ASK YOU WHAT DO YOU SEE? YOU SAY THEY LOOK LIKE A MESS AND IT WAS WIRES AND SO ON. SO IT DOESN'T HAVE REGULARITY WHEN YOU LOOK TAT LIKE THAT AND THIS IS NOT SATISFYING. BUT IF YOU PLOT IT AS A MATRIX IN THIS FORM, YOU SEE SOME REGULARITY. THIS IS THE SAME AS WHAT IS REPRESENTED NOW, THE SIZE AND COLOR OF THE SQUARES CORRELATED WITH THE PROBABILITY OF A CONNECTION. YOU SEE SOME REGULAR STRUCTURE. AND WHAT WE DID IN THE LAB, THEY CAME OUT WITH THREE RULES OR MOTIFS THAT RECAPITULATES THIS COMPLEX WIRING DIAGRAM YOU SEE HERE. I WILL TAKE YOU THROUGH THIS NOW AND THEN I WILL SHOW YOU THE QUANTIFICATION OF THE THE FIRST RULE IS THERE IS ONE TYPE OF INTERNEURON CALLED A MASTER REGULATOR. THAT IS THE MART KNOTTY CELL AND IT EXISTS IN LAYER 3 AND 5 AND EACH OF THESE CELLS WILL CONNECT TO EVERY OTHER SINGLE NEURON IN THE VICINITY. IT DOESN'T MATTER, IT WILL SHUT IT DOWN. BUT NO TOMB MART KNOTTY CELLS WILL CONNECT TO EACH OTHER. THEY DON'T TALK TO EACH OTHER. THIS IS THE FIRST RULE. THE SECOND RULE IS THE OPPOSITE. IT SEEMS LIKE THE CELLS IS THE OPPOSITE. THEY WILL ONLY CONNECT TO INTERNEURONS OF THEIR OWN TYPE AND THEY WILL NOT CONNECT TO ANY OTHER CELL EXCEPT PARAMEATALS. AND THE THIRD RULE IS THAT THERE IS A SISTER NEURONS THAT ARE NOT WELL-COBBLED INTO THE CIRCUIT. THEY DO NOT RECEIVE STRONG INPUT FROM THE LOCAL CIRCUIT AND WE SEE THOSE HAVE TO DO WITH BRAIN STATES AND MOD LATED BY CORTICAL AND OTHER STRUCTURES INCLUDING THE CHOLINERGIC SYSTEM. SO PUTTING THESE RULES TOGETHER TO BUILD A MODEL AND COMPUTED THE LIKELIHOOD OF THIS MODEL AND YOU CAN SEE THE QUANTIFICATION HERE. YOU SEE THESE THREE RULES, THIS IS THE REAL DATA AND THIS IS THE THREE RULES I DESCRIBED TO YOU AND YOU SEE HOW WELL YOU CAN GET THIS CONNECTIVITY DIAGRAM BY THESE THREE RULES. SO NOW THE GOAL IS TONED WHAT THESE RULES MEAN COMPUTATIONALLY IN TERMS OF THE FUNCTION. NOW IN PART OF WHAT WE ARE DOING, WE STARTED DOING THIS IN PRIMATES AND ALSO ABOUT TO DO IT IN HUMANS WHERE WE STUDIED THE MICROSURGERY OF THE PRIMATE AND WHAT IS INTERESTING IS NEW CELL TYPES AND NEW CONNECTIVITY PRINCIPLES THAT ONE DOESN'T FIND IN THE MOUSE PROVIDING THE EVOLUTIONARY PATH TO TRY TO UNDERSTAND HOW THE NEOCORTEX IS ORGANIZED. NOW, IN THE SUMMARY SO FAR, I DESCRIBED YOUR WORK SO FAR WHERE WE FOUND 15 SUBTYPES OF INTERNEURONS IN THE PRIMARY VISUAL CORTEX OF THE MOUSE. AND WE DEVELOPED THIS PATCH SEQ METHOD WHERE WE ARE COMBINING PATCHING WITH TRANSCRIPTOMIC ANALYSIS TO IDENTIFY THE TRANSCRIPTOME OF THESE CELLS. AND AS DESCRIBED TO YOU THREE CONNECTIVITY RULES THAT STATE HOW THE CORTICAL MICROCIRCUIT IS ORGANIZED AND THEN WE ARE NOW IN THE PROCESS OF THE FULL MICROCIRCUIT BY INCORPORATING L5 AND 6. I DIDN'T SHOW YOU SOME WORK DONE IN THE LAB WHERE SHE IS TRYING TO UNDERSTAND HOW THE PARAMEATAL CELLS ARE ORGANIZED AND LOOKING TAT FROM A DIFFERENT PERSPECTIVE. BUT I WON'T SHOW YOU THIS DATA BECAUSE I DON'T HAVE TIME. SO THE LAST PART OF THE TALK IS THE FUNCTION AND COMPUTATION WHICH IS WHAT MY LAB IS MAJOR FOCUS IS, AND ALSO THE HARDEST ONE TO GET TO. AND I WILL JUST DESCRIBE TO YOU APPROACH TO THIS AND SHOW YOU SOME ENCOURAGING RESULTS. WE ARE MOSTIVATED BY THIS IDEA IN ORDER FULLY UNDERSTAND COMPUTATIONALLY HOW THESE CORTICAL CIRCUITS WORK, WE HAVE TO BUILD LIKE A NETWORK OF NEURONS THAT IS MIMICKING WHAT THE HUMAN BRAIN ULTIMATELY DOES. SO WE ARE FOLLOWING A NICE CODE BY FINDING WHAT I CANNOT CREATE OR UNDERSTAND. SO OUR GOAL IS TO REVERSE ENGINEERING OF CORTICAL INTELLIGENCE AND ADVANCE MACHINE LEARNING IN THE SENSE WE WANT TO BUILD NEURONETWORKS THAT MIMIC WHAT WE OBSERVE IN ANIMALS OR EVENTUALLY HUMANS. THIS IS THE GOAL. AS I TOLD YOU BEFORE, RECENTLY THERE IS SCIENCE IN THIS FIELD AND IMPARTING BIG COMPANIES LIKE GOOGLE AND FACEBOOK BECOMING MORE NEURONETWORKS THAN ANYTHING ELSE AND AT THE HEART OF THIS PROBLEM IS YOU LEARNING NEURONETWORKS THAT OPERATE AND PERFORMANCE CANONICAL OPERATIONS WITH DIFFERENT WAYS OF COURSE YOU LEARN HERE AND WHAT WE BELIEVE IS THAT THESE DIFFERENT TYPES OF INTERNEURONS, THE WAY THEY CONNECT INTO THE CIRCUIT, THEY IMPLEMENT DIFFERENT LINNATORS. AND ONE OF OUR MAJOR GOALS IS TO UNDERSTAND WHAT THESE LINIATORY OPERATIONS ARE. NOW TO DO THIS, WE HAVE A COLLABORATION WITH MANY DIFFERENT INSTITUTIONS AND I SHOW THEM TO YOU AT THE END AND WHAT WE WANT TO DO IS COMBINE ANATOMY, INCLUDING HIGH-THROUGHPUT CONNECTOMICS USING PHYSIOLOGY BOTH IMAGING AND CUSTOM IMAGING AND ELECTROPHYSIOLOGY, MODELING AND MACHINE LEARNING TO CLOSE THE LOOP AND GO ALL THE WAY FROM BUILDING A CIRCUIT THAT IS PERFORMING SOMETHING. IN OTHER WORDS, TO REPLACE THE CURRENT NEURONETWORKS INTO A NEW FORM OF NEURONETWORKS THAT IS MORE INSPIRED. THE ULTIMATE EXPERIMENT WE WANT TO PERFORM AND THIS IS A PROJECT FUNDED, AND WE WANT TO RECORD AND IT'S BEEN INITIATED BY THE PIONEER AWARD REALLY, IS TO RECORD FROM A FULL VOLUME OF ONE MILLIMETER IN THE MOUSE OF ONE MOUSE, 100,000 NEURONS IN A PARTICULAR TASK, AND THEN ALSO DO A FULL ELECTROMICROSCOPIC CONSTRUCTION OF THE CORTICAL MICROCIRCUIT AND THEN FIND THE CELL TYPES BASED ON MORPHOLOGY, AND DO A FULL CONSTRUCTION WIRING DIAGRAM SO WE HAVE BOTH FUNCTION AND ANATOMY FOR THIS CORTICAL MICROCIRCUIT MODULE. AND THEN OBSTRUCT AND MODEL TO BUILD IT INTO A DEEP NEURONETWORK THAT IS GOING TO PERFORM CERTAIN TASKS. WE VALUES A PROJECT TO STUDY THE PLASTICITY RULES AND THIS IS A ONGOING THING THAT JUST STARTED UNDER THE BRAIN INITIATIVE FUNDED. FROM THE IMAGING POINT OF VIEW, WE ARE USING TWO PHOTON IMAGING WHICH ALLOWS US TO RECORD THE WHOLE CORTICAL MICROCIRCUIT OF ALL 100,000 NEURONS IN THIS CASE. AND TO DO THIS, WE HAVE TO DEVELOP NEW TECHNOLOGIES, ONE OF THE THINGS WE HAVE DONE, AND THIS WAS AGAIN FUNDED BY THE PIONEER AWARD S WE DEVELOPED A METHOD TO DO VERY FAST 3D DIMENSIONAL IN THE SMALL VOLUME OF CORTEX. YOU HAVE A MIRROR AND A -- UP AND DOWN AND HERE WE ARE USING DEFLECTORS TO STEER THE THREE DIMENSIONS VERY FAST. AND YOU CAN RECORD HIGH SPEEDS UP TO 500 NEURONS SIMULTANEOUSLY IN A SMALL BODY. ALSO COLLABORATION WITH CORNELL, PART OF OUR TEAM, WE ARE USING 3 NOT TON IMAGING TO IMAGE ALL THE WAY DOWN NON-INVASIVELY DOWN INTO THE HIPPOCAMPUS AND IN THIS CASE WE ARE IMAGING ALL SIX CORTICAL LAYERS AND BY LABELING INPUTS FROM DIFFERENT AREAS, WE CAN IMAGE THE FEEDBACK COMING FROM HIGH ORDER AREAS INTO OTHER AREAS LIKE SHOWN HERE BY AXONAL IMAGING. SO THE GOAL IS TO HAVE IN THIS SMALL VOLUME OF ABOUT 100,000 NEURONS IN THE MOUSE, BECAUSE IT WILL SPAN THREE VISUAL AREAS, WE WILL MAP INPUTS COMING FORWARD AND THE FEEDBACK AND ALL THE PROCESSES TAKING PLACE AND WE'LL HAVE THE WIRING DIAGRAM IN ORDER TO UNDERSTAND IT. NOW, WHAT WE HAVE STARTED DOING, WE STARTED BUILDING THIS KIND OF MODULES WHERE WE ARE USING OUR CURRENT DATA AND I WILL SHOW YOU THE PROGRESS SO FAR. SO WHAT WE HAVE DONE IS TRYING TO INSTEAD -- WHAT I HAVE SHOWN YOU EARLIER IS NEURAL NETWORKS THAT ARE START OF THE ART IN VISION, THEY ARE VERY SIMPLE. THEY DON'T HAVE ANY CELL TYPES FOR EXAMPLE AND HAVE A VERY SIMPLE NON-CANONICAL, NON LINEARITY. WE WANTED TO START PUT CONSTRAINTS BASED ON OUR EXPERIMENT IN PARTICULAR THAT I HAVE SHOWN YOU BASED ON THREE CONNECT ACTIVITY RULES AND THE CELL TYPES AND SEE WE COULD FURTHER ADVANCE AND ASK THE QUESTION AND BUILD A NEURONETWORK THAT IS CLOSER TO REAL BRAIN, REMEMBER BETTER IN TASKS. SO, THIS NEURONETWORK, WE HAVE THREE DESIGN PRINCIPLES. ONE ELEMENTARY BUILDING BLOCKS STARTING WITH THE CORTICAL MICROCIRCUIT. SO THEY HAVE CELL TYPES AND CANONICAL CONNECTIVITY IN THE MEASURES. THE OTHER IS THEY ARE FUNCTIONALLY, HIGH FUNCTIONAL FIDELITY IN THE SENSE WE ARE STUDYING SHOWING VISUAL STIMULATION TO THE MICE AND RECORDING ACTIVITY, TRYING TO FEED NEURONETWORKS SO THEY CAN BEST PREDICT ACTIVITY. AND MANY MODELS, THIS IS A VERY OLD FIELD IN VISION FOR EXAMPLE. PEOPLE HAVE BEEN BUILDING MODELS WHERE YOU BUILD A SIMPLE MODELING AND TRY TO PREDICT THE RESPONSES TO NATURAL IMAGES. AND THERE ARE A LOT OF PROMISE WITH THIS. CURRENTLY, WITH THIS NET MODULE, WHICH I'M SHOWING TO YOU HERE IS BASICALLY THREE MAJOR CELL TYPES AND IS WIRED IN A WAY THAT WE FOUND IN THE BRAIN BUT IS BUILDING IN A WAY THAT IT CAN BE TRAINED ON GPU AND SCALABLE AND LIKE BASICALLY A MACHINE LEARNING ALGORYTHM T CURRENTLY REACHED STATE-OF-THE-ART IN TERMS OF PREDICTING RESPONSES FROM THE MOUSE VISUAL CORTEX. WE EXTRACTED RINSE PELES FROM THIS LIKE NON LINNATORS LIKE NORMALIZATION AND OTHER THINS AND WE BUILD THEM INTO A NEURAL NETWORK AND BENCHMARK THEM AGAINST STANDARD TASKS AND THE ONES THAT PEOPLE USING MACHINE LEARNINGS IS SUPER RESOLUTION IMAGING. FOR EXAMPLE, YOU GIVEN A LOW RESOLUTION IMAGE AND YOU HAVE TO INFER THE HIGH RESOLUTION IMAGE FROM IT. THIS THEN COLLABORATION WITH OTHERS AS WELL. THE CURRENT NEURONETWORKS I HAVE SHOWN YOU, THIS IS THE HIGH RESOLUTION RECONSTRUCTED IMAGE AND IT IS MORE POPULARY AND THIS IS THE ONE -- THE PROJECTOR HERE DOESN'T SHOW YOU VERY WELL, BUT IF YOU -- BLURRY -- THIS IMAGE HAS A MUCH BETTER RESOLUTION AND IT IS SORT OF STATE-OF-THE-ART IN TERMS OF DOING SUPER RESOLUTION RECONSTRUCTION. THE DIFFERENT BETWEEN THESE NETWORKS IS THE ONE ON THE RIGHT HAS SOME THAT WE MEASURED IN THE BRAIN BRAIDS ON OUR EARLY STUDY OF CORTICAL MICROCIRCUITS. SO AS I TOLD YOU, THIS PROJECT IS VERY MULTIDISCIPLINARY AND IT INVOLVES EXPERIMENT COMPUTATION NEURO50s MACHINE LEARNING PEOPLE -- NEUROSCIENTISTS -- FROM STRUCTURE AND FUNCTION TO COMPUTATION. AND IT IS ALSO DONE IN COLLABORATION WITH THE ALAN INSTITUTE AND ALSO PRINCETON. I WOULD LIKE TO NOW JUST CLOSE BY ACKNOWLEDGING THE PEOPLE THAT DID THE WORK THAT I PRESENTED -- THE MORPHOLOGICAL WORK AND THE SEQUENCING WORK AND THE WIRING WORK. AND ALSO OTHERS FROM THE CAR LINS CAINSTITUTE WHO WE COLLABORATED AS WELL. THANK YOU. [ APPLAUSE ] >> THANK YOU. IS THERE QUESTIONS? WE HAVE TIME FOR ONE OR TWO IF YOU COME UP TO THE MICROPHONE, PLEASE. >> THANK YOU YOUR MASTER REGULATORS. SO, DO THEY NEVER TALK TO EACH OTHER OR THEY JUST END UP NOT TALKING TO EACH OTHER? >> THEY NEVER TALK TO EACH OTHER. >> HOW DO THEY KNOW THE OTHER ONE IS -- >> THIS IS INTERESTING. IT'S TRANSCRIPTOMICS. SO THERE IS SOMETHING SIMILAR WITH THE RETINA WITH MOSAIC PATTERNS AND SOME IDEA THAT LIKE -- OTHER MOLECULES THAT COULD SEND REPOLINGS. ONE IDEA IS THAT UNIQUE ID AND THIS UNIQUE ID ENABLES THEM TO RECOGNIZE EACH OTHER AND NEVER CONNECT. BUT THIS IS ALSO VERY STRIKING TO US THAT NEVER CONNECTIONS BETWEEN THESE CELLS AND EXTREMELY HIGH CONNECTIVITY PROBABILITY CLOSE TO 100% TO ALL OTHER NEURONS. SO -- >> YOU MADE A POINT WHEN YOU TALKED ABOUT THE RHESUS MONKEY IN CORTICAL WORK ABOUT PRIMATES AND ROLES AND DISTINCTIONS AND GENES AND CELL TYPES THAT ARE PRESSINENTS PRIMATES. A LOT OF FUNDAMENTAL WORK HAS BEEN DONE IN THAT AREA BUT MY QUESTIOTO YOU IS, WHEN YOU LOOK AT YOUR RNA-SEQ TRANSCRIPTOME DATE ARE YOU SEEING ANY PRIMATE-SPECIFIC GENES THAT DIFFERENTIALLY EXPRESS IN REGARDS TO CELL TYPES AND NETWORKS IN THE MORNINGY? >> YES, THIS IS SOMETHING WE ARE VERY INTERESTED IN PRIMATES. WE STARTED DOING PRIMARY WORKS IN TERMS OF MOREOVERROLOGY AND WIRING AND NOW STARTING TO DO THE TRANSCRIPTOMICS. THIS IS ONE OF THE THINGS BECAUSE OF THE INTERESTING IDEAS ABOUT CERTAIN GENES THAT ARE ESPECIALLY IN HUMANS IMPORTANT FOR EVOLUTION. BUT WE HAVEN'T LOOKED AT THOSE DATA YET. >> THERE IS DIVERSITY HOW THEY BEHAVE IN TERMS OF CORTICAL RHYTHM SO ON THE FAST AND SLOW-SPIKING CELLS. SEE THE MART KNOTTY CELLS, ARE THEY FAST OR SLOW? HOW DO YOU KIND OF THINK ABOUT INTEGRATING THESE DYNAMIC REGIMES INTO THIS MOREOVER OLLIE FRAMEWORK? >> VERY GOOD. I THINK THE PROBLEM IS PHYSIOLOGY, IT'S VERY INCOMPLETE THE WAY WE ARE DOING IT CURRENTLY IN THE WORLD, I MEAN. BECAUSE LIKE ESPECIALLY IN SLICES, YOU RUN SPECIFICALLY FROM A PROTOCOL THAT YOU LOOK AT THE FAST SPIKING AND THAT IS NOT WHAT ULTIMATELY IS THE FUNCTION IN THE COMPUTATIONAL CIRCUIT. IT'S NOT JUST LET'S SAY TO GENERATE THOUGHTS ACTIVITY DURING A PARTICULAR -- IT'S MUCH MORE SOPHISTICATED WE YET DO NOT UNDERSTAND. I DO NOT -- I THINK ULTIMATELY WE REALLY KNEW -- I THINK MORE LIKE WHEN YOU HAVE, YOU'RE TRYING TO LIKE SEE OPERATIONS THAT BUILD SOMETHING. DIFFERENT OPERATIONS BUT THEY ARE FUNCTIONAL OPERATIONS AND NOT CHARACTERIZED WITH JUST DOING THE PHYSIOLOGY WE AND OTHERS HAVE BEEN DOING. THAT'S WHAT I BELIEVE. I DON'T BELIEVE THERE ISN'T ONE. IT'S JUST WE DON'T KNOW HOW TO STUDY THAT RIGHT NOW. I'M SAYING THE CURRENT STANDARD METHODS THAT ONE USES GIVES YOU A AMBIGUITY IN TERMS OF CLASSIFICATION. NOW MORPHOLOGY AND COMPLETE BECAUSE MORPHOLOGY IS LIKE THAT'S IT. THERE IS NOTHING -- THAT THE LEVEL WHEN YOU SAY THE WELL-DEFINED PROBLEM, OTHER THAN SOME TECHNICAL DIFFICULTIES, YOU CAN COMPLETELY GET THE INFORMATION. AND I THINK THAT IS WHY MORPHOLOGY AND MORE OF A GHOST HERE. >> THANK YOU. >> THANK YOU VERY MUCH. I THINK WE ARE GOING TO MOVE ON. SO WE'LL SAVE QUESTIONS FOR THE BREAK. I'M GOING TO GO TO THE NEXT SPEAKER KEN SOLT, WHO HAS MULTIPLE PIS IN A TRANSFORMATIVE RESEARCH AWORD MIT, MGH AND HARVARD MEDICAL SCHOOL. WHO WILL BE TALKING ABOUT REDESIGNING, RECOVERY FROM GENERAL ANESTHESIA. >> I JUST TURNED THIS ON. CAN YOU HEAR ME OKAY? GREAT. SO, MY NAME IS KEN SOLT. I'M A CLINICIAN SCIENTIST AT MASSACHUSETTS GENERAL HOSPITAL AND HARVARD MEDICAL SCHOOL. ON BEHALF OF MY COLLABORATORS, EMERY BROWN, EDWARD BOYDEN AND MAT WILSON, I'D LIKE TO THANK THE MIFF FOR SUPPORTING OUR PROJECT. MY PARTICULAR GOAL IN THIS PROJECT WAS TO REDESIGN THE PROCESS OF RECOVERY FROM GENERAL ANESTHESIA AND AIDE LIKE TO SHARE WITH YOU SOME OF THE PROGRESS THAT WE HAVE MADE. SO IN CURRENT CLINICAL PRACTICE, THE WAY WE WAKE UP OUR PATIENTS FROM GENERAL ANESTHESIA IS ON AN ENTIRELY PASSIVE PROCESS. WHAT I MEAN IS THAT THE ONLY THING WE DO IS STOP DELIVERING THE GENERAL DRUG AND WE ESSENTIALLY ALLOW THE PHARMACOKINETICS OF DRUG CLEARANCE DICTATE WHEN OUR PATIENTS WAKE UP FROM GENERAL ANESTHESIA. YOU CAN MANAGE THIN ISN'T A VERY EFFICIENT WAY TO DO THINGS BUT IT'S REALLY THE ONLY WAY WE CAN DO THE THINGS RIGHT NOW. AND IT TURNS OUT THAT THERE ARE A NUMBER OF CLINICAL PROBLEMS RELATED TO THIS CLINICAL PARADIGM. MISJUDGMENT OF OF RESID YUM EFFECTS CAN LEAD TO SEVERE AIRWAY AND OXYGENATION PROBLEMS DURING EMERGENCE. SO WHEN YOU'RE TAKING A PATIENT FROM A VERY DEEP SURGICAL LEVEL OF ANESTHESIA WHERE THEY ARE UNCONSCIOUS AND NOT BREATHING AND YOU HAVE TO DECIDE WHEN YOU WILL PULL OUT THE BREATHING TUBE AND ALLOW THEM TO RECOVER. SOMETIMES IF YOU DON'T OR MISJUDGE HOW MUCH OF THE GENERAL ANESTHETIC IS STILL ONBOARD, YOU CAN GET INTO SERIOUS PROBLEMS AFTER PULLING OUT THE BREATHING TUBE. DILEAR YUM IS A COMMON PROBLEM IN PEDIATRIC ANESTHESIA. THE INCIDENTS CITED TO BE 20-30% IN PEDIATRIC PARENTS. SO THESE PATIENTS WAKE UP DELISHUOUS AND BECOME VERY DIFFICULT TO MANAGE. CURRENTLY WE HAVE NO WAY OF TREATING THIS PROBLEM OTHER THAN TO REANESTHETIZE THE PATIENT AND HOPE THEY WAKE UP BETTER THE SECOND TIME. AND INTERESTINGLY, HAVING THE PATIENT'S PARENTS PRESENT ON EMERGENCE DOESN'T CHANGE THE INCIDENTS AT ALL SUGGESTING THAT THIS HAS NOTHING TO DO WITH THE CHILD WAKING UP IN THIS FOREIGN HOSPITAL ENVIRONMENT BUT RATHER IT'S AN EFFECT OF THE GENERAL ANESTHETIC DRUG ON THE PEDIATRIC BRAIN. CAUSATIVE DYSFUNCTION IS GETTING MORE AND MORE PRESS. IT'S BEEN SHOWN THAT ELDERLY PATIENTS IS A HARDER TIME RECOVERING COGNITIVE FUNCTION AFTER GENERAL ANESTHESIA. THERE WAS A PAPER IN THE NEW ENGLAND JOURNAL A FEW YEARS AGO WHERE THEY FOUND IN PATIENTS WHO EXPERIENCED POSTOPERATIVE DILEAR YUM HIGHER INCIDENTS OF POSTOPERATIVE COGNITIVE DYSFUNCTION SUGGESTING THAT THE WAY THE PATIENTS WAKE UP FROM GENERAL ANESTHESIA IN THE SHORT-TERM CAN HAVE LONG-TERM CONSEQUENCES. AND AS I MENTIONED, THE WAY WE WAKE UP PATIENTS IS VERY INEFFICIENT. SO DELAYED EMERGE SENSE A COMMON PROBLEM WOO DEAL WITH EVERY DAY. SO SURGERY ENDS IN A VERY DISCRETE TIME POINTED BUT THE WAY WE WAKE UP THE PATIENTS IS DICTATED BY PHARMACOKINETICS. IF YOU MISTIME THAT, SOMETIMES YOU MIGHT BE WAITING FOR 10-15 OR 20 MINUTES FOR PATIENTS TO WAKE UP AND THIS IS WASTED TIME IN THE OPERATING ROOM WHERE NOTHING CAN HAPPEN TO START GETTING THE NEXT CASE UNDERWAY AND THIS ESTIMATE OF ABOUT 1000 DOLLARS AN HOUR WAS A CONSERVATIVE ESTIMATE THAT IS ABOUT 10 YEARS OLD NOW. SO THE COST OF BEING IN A OPERATING ROOM IS VERY EXPENSIVE. IN OUR HOSPITAL WE HAVE 70 OPERATING ROOMS, EVEN A 5 OR 10 MINUTE DELAY IN A WAKEUP CAN ADD UP IN A SINGLE DAY IN TERMS OF COST. OUR GROUP HAS BEEN WORKING ON THIS PROBLEM OF TRYING TO TAKE EMERGENCE FROM GENERAL ANESTHESIA AND TURNING IT FROM A PASSIVE PROCESS FROM SHUTTING OFF DRUGS TO ACTIVE PROCESS WHERE WE PERFORM INTERVENTION TO PULL THE PATIENT OUT OF THE ANESTHETIZED STATE BACK TO THE AWAKE STATE. AND IT TURNS OUT THAT GENERAL ANESTHETICS ARE VERY PROMISCUOUS DRUGS AND HIT A LOT OF TARGETS AND RECEPTORS ET CETERA. AND SO, A MOLECULAR ANTAGONISM APPROACH TO REVERSING GENERAL ANESTHESIA ISN'T FEASIBLE BECAUSE OF THE NATURE OF THE DRUGS WE ARE DEALING WITH. AND SO WE TOOK AN ALTERNATIVE APPROACH TO THIS PROBLEM BY LOOKING AT THE AROUSAL CIRCUITS, SUBCORTICAL AROUSAL CIRCUITS IDENTIFIED OVER THE YEARS BY SLEEP BIOLOGISTS. AS YOU CAN SEE, THERE ARE A NUMBER OF AROUSAL PROMOTING NEUROTRANSMITTERS IN THE BRAIN. ALL OF THESE PROMOTE AROUSAL IN THE BRAIN. FROM IS ALSO ANOTHER PEPTIDE NEUROTRAPS MITTER RELEASED BY THE HYPOTHALAMUS. SO WE BECAME INTERESTED IN LOOKING AT WAYS TO STIMULATE THESE AROUSAL CIRCUITS IN THE BRAIN TO REVERSE THE STATE OF GENERAL ANESTHESIA. THE ONES OR THE NEUROTRAPS MITTERS WE FOCUSED ON SPECIFICALLY WERE DOPAMINE AND NOREPINEPHRINE. THE REASON IS, CLINICALLY AVAILABLE STIMULANT DRUGS, RITALIN, USED TO TREAT ADHD, IT TURNS OUT THAT IT IS REUPTAKE INHIBITOR FOR DOPAMINE AND NOREPINEPHRINE. SO WE HYPOTHESIZED SEVERAL YEARS AGO THAT PERHAPS A DRUG LIKE THIS COULD BE USED TO REVERSE THE STATE OF GENERAL ANESTHESIA. AND WE FOUND THIS WAS THE CASE. SO 2011, WE PUBLISHED A PAPER SHOWING THAT THIS REVERSES THE STATE OF GENERAL ANESTHESIA IN RODENTS. WE FOUND THAT IT REVERSED VERY DISTINCT GENERAL AB THREATICS SUGGESTING THE HYPOTHESIS WAS CORRECT. -- ANESTHETICS. AS A CIRCUIT LEVEL ANTAGONIST BY STIMULATING AROUSAL WITH THE SUB-CORTICAL CIRCUITS. SO THEN WE FOLLOWED UP WITH SEVERAL STUDIES TO TRY TO FIGURE OUT WHETHER THE EFFECT OF THIS WAS DRIVEN BY DOPAMINE OR NOREPINEPHRINE IN THE BRAIN. IT'S REUPTAKE INHIBITOR FOR BOTH AND AS I SHOWED YOU IN THE PREVIOUS SLIDE, BOTH ARE INVOLVED IN BEHAVIORAL LEARNING. SO BY JUST GIVING METHYL FEN DATE WE DIDN'T FINISH THIS WAS DRIVEN BY DOPAMINE OR NOREPINEPHRINE. THE MAIN NEUROTRANSMITTER INVOLVED WAS DOPAMINE. AND SO THE NEXT QUESTION WE ASKED IS WHERE IS THE DOPAMINE COMING FROM IN THE BRAIN? SO DOPAMINE IS NOT REALLY BEEN WELL CHARACTERIZED AS AROWDAL PROMOTING TRANSMITTER IN THE SLEEP NEUROBIOLOGY COMMUNITY. IT TERPS OUT THAT DOPAMINE NEURONS TONED HAVE STABLE FIRING RANGES AND THAT LARGELY MADE SLEEP BIOLOGISTS DISINTERESTED IN STUDYING THIS. BUT WE HAD A STRONG FEELING THAT IT HAD AROUSAL EFFECT THAT COULD REVERSE GENERAL ANESTHESIA BUT DIDN'T KNOW WHERE THE DOPAMINE WAS COMING FROM. WHEN WE GIVE A DRUG, LIKE A REUPTAKE INHIBITOR, WE ARE NOT SUPPLYING A SOURCE OF DOPAMINE SO IT HAS TO BE COMING FROM SOMEWHERE IN THE BRAIN. THE TWO MAIN SOURCES OF DOPAMINE IN THE BRAIN ARE HERE. THE SUBSTANCE NIAGRA NEURONS ARE HEAVILY STUDIED IN PARKINSON'S AND CRITICALLY INVOLVED IN MOVEMENTS WHEREAS THE OTHER ARE HEAVILY STUDIED IN THE CONTEXT OF REWARD AND THOUGHT TO BE A REWARD PATHWAY IN THE BRAIN. WHICH OF NIECE NEURONS PROMOTES AROUSAL WASN'T CLEAR. SO WE PERFORMED A STUDY USING ELECTRICAL STIMULATION OF THE VENTRAL AREA AND THE SUBSTANTIAL NIAGARA TO FIGURE OUT IF ELECTORAL STIMULATING THESE PARTS OF THE BRAIN WILL -- WE FOUND THAT RATS WOULD WAKE UP AND FLIP OVERAND ACT CONSCIOUS WHEN WE STIMULATED THE VENTRAL TAG MENTAL AREA BUT IT DIDN'T OCCUR WHEN WE STIMULATED THE SUN STANCE NIGHTA SUGGESTING THE DOPAMINE REVERSING THE STATE IS COMING FROM THE VTA AND NOT THE SUBSTANCE NIAGRA. HOWEVER, THE PROBLEM WAS WITH THAT KIND OF APPROACH, WE COULDN'T PROVE THAT DOPAMINE WAS DRIVING THIS EFFECT. WE USED ELECTRICAL STIMULATION, WHEREVER THAT ELECTRODE IS, IT WILL STIMULATE THE NEURONS IN THAT AREA, SO THE VTA, EVEN THOUGH IT'S OCCURRED A DOPAMINE CENTER IN THE BRAIN, HAS BOTH DOPAMINE NEURONS AND NON-DOPAMINE NEURONS. WHEN YOU USE ELECTRICAL STIMULATION IT WILL ACTIVATE ALL CELL TYPES IN THE AREA OF THAT ELECTRODE TIP SO YOU CAN'T REAL YOU PROVE THE EFFECT IS DRIVEN BY DOPAMINE. WE WERE ABLE TO USE NEUROTECHNOLOGY WITH OPT GENETICS IN COLLABORATION WITH ED BOYDEN OUR COLLEAGUE AT MIT; TO SELECTIVELY TARGET DOPAMINE NEURONS USING CHANNEL REDOPS IN, AND A GENETIC APPROACH TO TARGET THIS IN DOPAMINE NEURONS SELECTIVELY AND THEN WE USED A BLUE LASER LATE TO ACT VITTHOSE NEURONS TO SEE IF WE CAN REVERSE THE STATE OF GENERAL ANESTHESIA BY NORM TAILOR IN MY GROUP AND THIS IS BASICALLY THE GENETIC APPROACH TO TARGET JUST THE DOPAMINE NEURONS WITH THIS LIGHT SENSITIVE ION CHANNEL. SO WE HAVE THE GENE ENCODING GENERAL KIDOPS IN, HAVE IT IN A VIRUS AND WE INJECTED IT INTO THE VTA AREA OF DAT CRE MICE, GENETICALLY MODIFIED TO EXPRESS CRE RECOMBINASE UNDER THE CONTROL OF THE DOPAMINE TRANSPORTER. SO IT'S ONLY EXPRESSED IN DOPAMINE NEURONS IN THE GENETLY MODIFIED MICE. SO WHEN WE INJECTED VTA WITH THE VIRUS THAT ENCODES THIS FLOCKS OF GENES TO EXPRESS GENERALRI DOPS IN, THE VIRUS MAY TRANSFECT NON DOPAMINE NEURONS IN THE AREA HOWEVER, WITHOUT CRE IN THE CELL, IT WILL NOT EXPRESS RHODOPSIN. SO THE AREA IS WE INJECT THIS VIRUS INTO THE VTA BUT ONLY THE DOPAMINE NEURONS WILL EXPRESS THIS CAT ION CHANNEL THAT OPENS IN RESPONSE TO BLUE LIGHT AND DEPOLARIZES THE NEURONS. THIS IS THE APPROACH THAT NORM USED TO ACTIVATE THE DOPAMINE NEURONS SELECTIVELY UNDER GENERAL ANESTHESIA AND THIS IS A VIDEO OF ONE OF HIS EXPERIMENTS. SO THIS IS A MOUSE UNDER ANESTHESIA, A DAT CRE MOUSE THAT UNDERWENT THE PROCEDURE I SHOWED IN THE PREVIOUS SLIDE. HE HAS FIBER OPTICS IN THE VTA AND HE IS BREATHING INHALING THIS AND YOU'LL SEE A BLUE LIGHT COME ON IN JUST A MOMENT. AND NOW THAT BLUE LIGHT IS OPENING UP THOSE ION CHANNELS, DEPOLARIZING DOPAMINE NEURONS IN THE VTA AND AS YOU CAN SEE, AS WE CONTINUE TO STIMULATE THE ANIMAL STARTS TO WAKE UP FROM THE ANESTHETIC. THE KEY POINT HERE IS THAT THE ISOFLUORINE IS ALWAYS THERE. WE HAVE NOT TURNED OFF OR TURNED DOWN THE ANESTHETIC. JUST BY ACTIVATING THE DOPAMINE NEURONS IN THE VTA, WE ARE ABLE TO GET THIS MOUSE TO WAKE UP. WE USED THE RIGHTING REFLEX, THE ANIMAL FLIPPING TO ALL FOURS TO INDICATE RETURN OF CONSCIOUSNESS. WE DIDN'T MAKE IT EASY WITH THIS IMPLANT IN THE HEAD BUT AS YOU CAN SEE, THE ANIMAL DID FLIP OVER AND EXHIBITING CONSCIOUS BEHAVIORS SUGGESTING THE ANIMAL HAS WOKEN UP FROM ANESTHESIA AND THE STATE HAS BEEN REVERSED. BUT THENY QUESTION IS, HAVE WE RESTORED COGNITIVE FUNCTION? ONE COULD ARGUE WE JUST RESTORED AROUSAL BUT IT DOESN'T PROVE WE WOKEN UP THE ANIMAL AND RESTORED COGNITIVE FUNCTION. THAT WAS THE NEXT STEP THAT WE WANTED TO TAKE IN RESEARCH TO TRY TO SEE IF THESE MANIPULATIONS WOULD RESTORE COGNITIVE FUNCTION IN THESE ANIMALS AND REVERSE ANESTHESIA IN THAT WAY. AND THESE ARE THE COGNITIVE TESTING CHAMBERS THAT WE NOW HAVE IN OUR LABORATORY. EACH ONE IS FOR A SINGLE ANIMAL. AND WHEN YOU OPEN UP THESE CHAMBERS, THIS IS WHAT IT LOOKS LIKE ON THE INSIDE. THERE IS A TOUCH SCREEN ON THE LEFT AND THERE IS A FOOD DISPENSER ON THE RIGHT. AND SO WHEN THE ANIMAL PERFORMANCE A VISUAL DISCRIMINATION TASK, THE ANIMALS THAT WE USED IN THIS STUDY ARE RATS BECAUSE THEY ARE A LITTLE BIT BIGGER AND EASIER TO MANIPULATE AND TEACH. BUT WHAT WE HAVE DONE IS TAUGHT THESE ANIMALS TO SELECT THIS IMAGE, THE DOTS, OVER THIS IMAGE, FOR FOOD REWARD AND ARE IT TAKES 6-8 WEEKS TO GET THEM FULLY TRAINED UP TO PERFORM THIS TASK. BUT THEY PERFORM PRETTY WELL. SO IF THEY TOUCH THIS IMAGE, A FOOD PELLET WILL DROP BACK HERE AND THEN THEY'LL REPEAT THAT. AND I'LL SHOW WHAT YOU THAT LOOKS LIKE. WE CAN RUN 4 AT A TIME. SO HERE IS ANIMAL HERE THAT JUST INITIATED A TRIAL. THIS IS THE CORRECT IMAGE OVER HERE AND YOU'LL SEE HE'LL TOUCH THAT IMAGE AND GO BACK AND GET HIS FOOD. THIS ANIMAL IS ABOUT TO RUN A TRIAL AS WELL. HE COMES OVER AND THINK ABOUT THIS ONE. KNOWS THAT IS THE WRONG ONE AND THEN HITS THE RIGHT ONE. THEY GET 95% ACCURACY IN 6-8 WEEKS. THEY BECOME VERY GOOD AT PERFORMING THIS TEST FOR FOOD REWARD AND SO THEN WE ASK OURSELVES A QUESTION. IF YOU TAKE ANIMAL WHO HAS BEEN TRAINED TO DO THIS TASK AND ANNETTE TIES THEM, CAN WE RESTORE THE ABILITY TO DO THIS TASK BY STIMULATING VTA? THIS IS AN IMPLANT IN THE VTA. THE WHOLE CHAMBER IS INFUSED WITH 1/2% ISOFLUORINE, A PRETTY SIGNIFICANT SEDATIVE DOSE. NOT QUITE ANESTHETIC DOSE. JUST TURNED ON THE VTA STIMULATION WITHIN A COUPLE OF SECONDS, THE ANIMAL WAKES UP AND STARTS PERFORMING THE TASK. AND EVEN THOUGH THE ISOFLUORINE IS THERE THE WHOLE TIME, THIS WAS A DOSE THAT BASICALLY MADE ALL OF OUR ANIMALS UNABLE TO PERFORM THIS TASK BY STIMULATING VTA UNDER THESE CONDITIONS WE WERE ABLE TO RESTORE THE ABILITY TO PERFORM THIS EARLY COMPLEX COGNITIVE TASK SO THEY ARE NOT JUST FLIPPING OVER, THEY ARE -- WE ARE TURNING THEIR BRAINS BACK ON. HE DOES GET MORE OF THEM WRONG, HOWEVER. WE ARE STILL ANALYZING THIS DATA. HE IS GOING TO MAKE ONE MISTAKE. THE LIGHT COMES ON IN HIS FACE AND IT'S A MILDLY ADVERSIVE STIMULANT. I WANT YOU TO SEE WHAT HAPPENS WHEN WE TURN OFF THE STIMULANT. THE ANESTHETIC IS THERE THE WHOLE TIME AND WE ARE ABOUT TO TURN THE STIMULATION HERE AND EVEN THOUGH HE IS COMING OVER TO DO THIS TASK, JUST BY TURNING OFF THE STIMULATION HE GOES RIGHT BACK DOWN. THE ANESTHESIA IS THERE THE WHOLE TIME. SO THE VTA STIMULATION ALMOST LIKE A LIGHT SWITCH. WE ARE ABLE TO TURN IT ON AND OFF AT WILL BY ACTIVATING THE VTA. WE HAVE BEEN USING ELECTRICAL STIMULATION FOR STUDIES BECAUSE OF THE COMPLEXITY OF THE TASK AND EVERYTHING WE ARE DOING BUT HOPING TO MOVE FORWARD WITH MORE SELECTIVE TARGETING TECHNIQUES BUT WE ARE PRETTY EXCITED ABOUT THESE RESULTS. AND SO JUST TO SUMMARIZE WHAT I HAVE SHOWN YOU TODAY, I HAVE SHOWN YOU THAT OPT GENETIC STIMULATION OF VTA DOPAMINE NEURONS INDUCES REANIMATION FROM GENERAL ANESTHESIA. SO INCREASE IN DOPAMINERGIC TRANSMISSION MAY BE A TRIGGER THAT DRIVES THE PROCESS OF WAKING UP FROM ANESTHESIA EVEN THOUGH IT MAY NOT PLAY A KEY ROLE IN SLEEP-WAKE TRANSITIONS, IT MAY BE PLAYING A CRITICAL ROLE IN ANESTHESIA TO AWAKE, TRANSITION. WE ALSO FOUND THAT DOPAMINERGIC AGENTS AND OTHER CLINICALLY AVAILABLE STIMULANT DRUGS MAYBE USEFUL TO INDUCE REANIMATION. WE ARE PERFORMING A CLINICAL TRIAL RIGHT NOW TO TEST THAT VERY HYPOTHESIS AND TO SEE IF PATIENTS WAKING UP FROM ANESTHESIA WAKE UP QUICKER AND BETTER WITH A DRUG LIKE METHYL FEN DATE WHICH WE GIVE TO PEOPLE ALL THE TIME. SO WE ARE DOING A CLINICAL TRIAL WITH THAT NOW. AND FINALLY, ELECTRICAL VTA STIMULATION PROMOTES AROUSEAL AND RESTORES COGNITIVE FUNCTION IN THE RODENTS. SO WE FEEL THAT EVEN THOUGH THIS MAY BE QUITE AWAYS AWAY, THIS SORT OF APPROACH MAY BE USEFUL TO RESTORE AROUSAL AND COGNITIVE FUNCTION IN PATIENTS SUFFERING FROM DISORDERS OF CONSCIOUSNESS. BECAUSE GENERAL ANESTHESIA IS MORE AKIN TO A DRUG-INDUCED COMA THAN TO SLEEP. WE FEEL IS THERE A POSSIBILITY THAT WE COULD START TREATING PATIENTS WITH TRAUMATIC BRAIN INJURIES, NEURODEGENERATIVE DISORDERS, WHO OBVIOUSLY YOU WOULDN'T USE BRAIN EMANANT WAKE UP A PATIENT FROM ANESTHESIA BUT TO WAKE UP A PATIENT SUFFERING FROM DISORDERS OF CONSCIOUSNESS, THIS MAY BE A VERY REAL TECHNIQUE MAYBE USED TO HELP THOSE PATIENTS. SO, I LIKE TO JUST WRAP UP THERE AND THANK MY COLLABORATORS AND MY LAB MEMBERS WHO DID MOST OF THE WORK AND THANK YOU FOR YOUR ATTENTION AND I'M HAPPY TO TAKE ANY QUESTIONS. [ APPLAUSE ] >> HI. YOU MENTIONED THAT THE PROBLEM OF DILEAR YUM IS MORE COMMON AMONG CHILDREN AND I WAS WONDERING IF YOU THINK YOUR RODENT MODEL IS MORE A MODEL OF ADULTS VERSUS CHILDREN AND HOW THAT -- HOW MIGHT YOU GET AT THE ISSUE THAT MIGHT BE MORE SPECIFIC TO CHILDREN? >> SO THAT'S A GREAT QUESTION. I THINK SO ALL OF THE RODENTS WE USE WERE ADULT RODENTS IN OUR STUDY. AND SO, WE REALLY DON'T KNOW WHY CHILDREN ARE SO MUCH MORE SUSCEPTIBLE TO THIS DILEAR YUM IT'S SOMETHING LIKE A HUGE PROBLEM IN PEDIATRIC ANESTHESIA. WE DON'T REALLY SEE IS THAT MUCH IN ADULTS. YOU SEE IT A LITTLE BIT MORE WITH ELDERLY PATIENTS SO THE EXTREMES ARE THERE MORE. BUT OUR MODEL IS FOCUSED MORE ON THIS -- SORT OF ADULT POPULATION TO TRY TO SEE IF THESE THINGS WILL WORK. I AM COLLABORATING WITH A PEDIATRIC ANESTHESIOLOGIST TO SEE IF A DRUG LIKE METHYL FEN DATE COULD ACTUALLY MITIGATE OR EVEN TREAT EMERGENCE DILER GLUM CHILDREN. WE REASON ANESTHETIZE THE KIDS. THEY ARE THRASHING AROUND AND PULLING OUT IVs AND DOING STUFF THAT IS NOT SAFE. OUR APOACH IS TO GIVE THEM ANESTHETIC AGAIN AND PUT THEM BACK UNDER AND THEN HOPE THEY WAKE UP BETTER THE NEXT TIME. SO IT'S NOT A GREAT WAY TO DO THINGS. IF YOU CAN PULL THEM OUT OF THAT STATE IT WOULD BE GREAT. WE THINK SOME OF THESE APPROACHING MAY HELP FOR THAT. >> I WAS INTERESTED IN THE FACT THAT YOU CAN'T, IF I UNDERSTAND RIGHT, THE GENERAL ANESTHETIC DRUG BEING ADMINISTERED WHILE STIMULATING VTA. AND I WONDERING WHAT ARE THE IMPLICATIONS -- FIRST THAT ISN'T WHAT YOU WOULD DO CLINICALLY. AND BUT MORE INTERESTINGLY, WHAT DOES THAT TELL US ABOUT WHETHER THOSE TWO SYSTEM, THE AROUSAL AND WHATEVER IS HAPPENING WITH ANESTHESIA ARE INHIBITING WITH EACH OTHER AND RATHER THAN BEING PARALLEL SYSTEMS? >> IT'S A GREAT QUESTION. RIGHT NOW, WE KNOW A LOT ABOUT THE MOLECULAR MECHANISMS OF ANESTHETICS AND HOW THEY EFFECT GAB RECEPTORS IN THE BRAIN BUT WE ARE REALLY SORT OF VERY EARLY IN OUR UNDERSTANDING OF HOW THESE DRUGS EFFECT NEUROCIRCUITS AND HOW THE MOLECULAR EFFECTS TRANSLATE TO EFFECTS AT THE LEVEL OF CIRCUITS. THE REASON WHY I DID THE EXPERIMENT THE WAY I DID WAS TO BASICALLY MAKE IT UNEQUIVOCAL TO CONTROL THE ANIMAL. WE CAN BASICALLY DIAL IN A LEVEL OF SEDATION WITH GENERAL ANESTHESIA. SO WHAT I WANTED TO SHOW IS WHILE THE ANIMAL IS UNDER STEADY STATE, CAN WE REVERSE THAT STATE? WE CAN ALSO WAKE THEM UP IN THE OPERATING PROGRAM AND SEE IF THAT PROCESS GOES FASTER BUT THERE ARE OTHER FACTORS IN PLAY WHEN WE DO THAT. SO INHALED ANESTHETICS TENDS TO BE CLEARED FASTER WHEN YOU BREATHE FASTER. SO, IF SOME OF OUR INTERVENTIONS MAKE THE ANIMAL BREATHE FASTER, THAT COULD ACCOUNT BY ITSELF FOR THE FASTER WAKE UP. SO JUST TAKE ALL THOSE COMPOUNDING VARIABLES OUT OF THE WAY, AND THESE KIND OF STUDIES I HAVE BEEN DOING IT JUST DOING UNDER CONTINUOUS ANESTHESIA. BUT WE DON'T KNOW YET HOW ANESTHETICS EFFECT THE AROUSAL CIRCUITS. WHAT IS HAPPENING DURING PASSIVE EMERGENCE OR WHERE WE ARE STIMULATING CIRCUITS TO WAKE UP THE ANIMAL. I THINK IS THERE IS A LOT MORE WE NEED TO LEARN ABOUT HOW THE SYSTEMS WORK. >> HI. I'M CURIOUS IF YOU TRIED ANY OTHER AROUSAL PROMOTING CENTERS? SO THE HIST MA NERGIC NEURONS HAVE THE BEST CORRELATION BETWEEN FIRING PATTERNS AND WAKE FULLNESS, PART OF NORMAL SLEEP WAKE CIRCUITS BUT AS ARE THE VTA NEURONS AS YOU PUBLISHED A FEW WEEKS AGO IN THAT BEAUTIFUL PAPER. SO I'M CURIOUS, IS THIS SOMETHING THAT YOU COULD GET WITH ANY AROUSAL CENTER IN THE BRAIN MANIPULATION? >> THAT'S A GREAT QUESTION. WE PLAYED AROUND WITH OTHER SYSTEMS SO WE FOUND THAT ACTIVATING -- MOST OF THESE STUDIES ARE WITH EXOGENOUS DRUGS SO WHEN WE GAVE DRUGS WHICH ACTIVATE CHOLINERGIC CENTER WE DIDN'T GET THIS KIND OF RESPONSE. WHEN WE GAVE ADDA MOCKS TEEN, STRA TARA USED FOR ADHD, A NOREPINEPHRINE REUPTAKE INHIBITOR, THAT DIDN'T WAKE THEM UP EITHER. THERE IS SOMETHING ABOUT DOPAMINE THAT ALLOWS US TO REVERSE ANESTHESIA. HISTAMINE HAS BEEN USED BUT IT'S A SOFT EFFECT WHEN IT COMES TO THIS. IF YOU THINK WITH THE FACT WHEN YOU TAKE A DRUG LIKE BENADRYL WHICH MAKES YOU SLEEPY, IT PRESUMABLY ACTS BY BLOCKING HISTAMINE WHICH AROUSE PROMOTING IN THE BRAIN BUT IT DOESN'T PUT YOU UNDER GENERAL ANESTHESIA. IT'S NOT THAT DEEP. WE NEED TO START TEASING APART DIFFERENCES BETWEEN SLEEP, THE CIRCUITS INVOLVED IN SLEEP AND AROUSAL AND ANESTHESIA. >> BUT EXCELLENCING VTA NEURONS ALSO DOESN'T AUTOMATICALLY PUT ANIMALS INTO KIND OF COMA. THEY GO AND MAKE A NEST. AND THEN STIMULATION OF HIST MA NERGIC NEURONS MAY BE DIFFERENT THAN GIVING -- >> I THINK WHAT WE NEED TO DO IS START TEASING APART ALL THE CIRCUITS. AND WE ARE AGAIN NOT THERE YET. THIS IS JUST THE -- THERE ARE OTHER GROUPS WORKING ON OTHER CIRCUITS. BUT NOTHING THAT HAS BEEN DRAMATIC AS THIS EFFECT WE HAVE SEEN. >> LAST QUICK ONE. >> ONE OF THE PREVIOUS QUESTIONERS ASKED WHY WE WANT TO STIMULATE IN ANIMAL WHO HAS ISOFLUORINE ONBOARD. I HEARD AN INTERESTING SEMINAR BY JIM SUMINAR LAST WEEK WHO IS AN ANESTHESIOLOGIST THAT POINTS OUT SOME METABOLIC SYNDROMES PRODUCE ANESTHETIC-LIKE MOLECULES THAT PUT PATIENTS INTO COMAS. SO THE IDEA WOULD BE, COULD YOU WAKE UP ONE OF THESE PATIENTS THAT IS IN THAT METABOLIC DISEASE-INDUCED COMA WITH YOUR TECHNIQUES? >> THANK YOU. I THINK THE THING ABOUT ANESTHESIA IS IT IS CONSISTENT FROM ANIMAL TO ANIMAL. THE THING WITH METABOLIC CAUSES OR NEURODEGENERATIVE CAUSES OR TRAUMATIC BRAIN CAUSES OF DISORDERS OF CONSCIOUSNESS IS THAT THE DISORDERS ARE HETEROGENOUS, I THINK. SO, GETTING TO EACH OF THOSE IS GOING TO BE TRICKY. IT MAY WORK FOR SOME BUT NOT OTHERS. THESE ARE ALSO THINGS WE ARE INTERESTED IN COLLABORATING WITH NEUROLOGISTS TRY GET TO SOME OF THESE QUESTIONS. CAN WE PREDICT IF YOU SEE DAMAGE IN A SPECIFIC PART OF THE BRAIN, WHETHER A PATIENT WOULD WAKE UP WITH SOME OF THESE TECHNIQUES I THINK IT WILL BE CIRCUIT-DEPENDENT, THAT YOU MAY NEED TO TARGET OTHER CIRCUITS DEPENDING ON WHAT THE DAMAGE IN THE BRAIN IS. BUT I THINK IT'S AN INTERESTING QUESTION TO START PURSUING. >> THANK YOU VERY MUCH. >> THANK YOU. [ APPLAUSE ] >> SO OUR FINAL SPEAKER IF FOR THIS SESSION IS ALAN ANTICEVIC OF YALE UNIVERSITY. ALAN IS ONE OF OUR EARLY INDEPENDENCE AWARDEES. HE IS GOING TO TELL US ABOUT MECHANISTICALLY INFORMED NEUROIMAGING MARKERS FOR MENTAL ILLNESS USING PHARMACOLOGY AND COMPUTATION. >> I WANTED TO MAKE SURE THAT EVERYBODY CAN HEAR ME OKAY. GOOD. SO THANK YOU VERY MUCH FOR THE OPPORTUNITY TO GIVE A TALK AND TO PRESENT WHAT WE HAVE BEEN UP TO. SO, I'D LIKE TO GET ACKNOWLEDGMENTS FIRST SO I DON'T FORGET. THIS IS REALLY MULTIDISCIPLINARY WORK THAT INVOLVES MANY, MANY PEOPLE. BY NO MEANS JUST DONE BY ME. AND A NUMBER OF IMPORTANT COLLABORATORS, COLLEAGUES, GRAD STUDENTS AND I WANTED TO PARTICULAR THANK THE NIH DIRECTORS EARLY INDEPENDENCE AWARD FUNDING MECHANISM FOR LAUNCHING US AND FOR ALLOWING US TO DO THIS WORK. SO WITH THAT SAID, I WANT TO SHOW A PICTURE THAT LOOKS COMPLICATED BUT MAKES A SIMPLE POINT. WHAT YOU'RE LOOKING AT ON THE LEFT IS MEN AND ON THE RIGHT IS WOMEN. THE X AXIS IS AGE AND THE Y AXE SIS DAYS LOST DISABILITY. AND THIS BURDEN OF MENTAL ILLNESS ACROSS THE LIFESPAN. SO I HOPE YOU CAN SEE THAT IT STRIKES EARLY IN LIFE, IT PERSISTS, AND IT REMAINS PRETTY MUCH UNTREATED. SO IT'S A MAJOR SOURCE OF DISABILITY AS WELL AS ECONOMIC HARDSHIP FOR MOST PEOPLE WHO DEVELOP IT. SCHIZOPHRENIA IN PARTICULAR, WHEN I'LL TALK ABOUT TODAY, IS TOP 10 LEADING CAUSE OF DISABILITY IN THE WORLD. PEOPLE WHO GET IT TEND TO DIE ABOUT 25 YEARS YOUNGER DUE TO EITHER SUICIDE OR VARIOUS MEDICAL COMPLICATIONS. AND IT'S FAIR TO SAY WE REALLY DON'T HAVE NEW TREATMENTS FOR THIS DISORDER ON THE OTHER HAND SERENDIPITOUSLY DISCOVERED THINGS AND THERE ARE STILL SYMPTOMS THAT COMPLETELY REMAIN OUT OF REACH FOR CURRENT TREATMENT MODALITIES. SO MAYBE IT'S A PLATITUDE BUT I HOPE WE CAN AGREE IN ORDER TO FIX THIS PROBLEM, WE NEED TO FULLY UNDERSTAND THE MECHANISMS THAT ARE DRIVING THESE SYMPTOMS. SO THE CHALLENGE THAT I HOPE TO ILLUSTRATE IN THIS CARTOON IS TO BRIDGE PSYCHIATRIC SYMPTOMS TO FUNCTIONALITY OF NEURONAL SYMPTOMS AND THE FUNCTIONALITY OF SINGLE CELLS AND CIRCUITS THAT UNDRESS CELL ABILITIES AND PERHAPS IT'S CONTROVERSY TOOL SAY BUT WE DON'T HAVA I UNIFYING INTERESTING FOR A SINGLE COMPLEX PSYCHIATRIC SYMPTOM IN 2016. WE UNDERSTAND FEAR CIRCUITS VERY WELL BUT STILL CAN'T TREAT PTSD UNIFORMLY. SO WHAT DO WE DO? MY LAB COMES AT THE PROBLEM AT THE LEVEL OF NEURAL SYSTEM. SO CLINICAL IMAGER WHO STUDIED THE RELATIONSHIP BETWEEN LARGE-SCALE BRAIN SYSTEMS ANDATISM THOMAS. AND THAT IS GREAT BUT IT LEAVES A VAST EXPLANATORY GAP OF LINKING RICH AND DETAILED BRAIN PICTURES TO UNDERLINE BRAIN MECHANISMS. ONE APPROACH WE CAN HARN SAYS TO START FROM MATHEMATICALLY BASED MODELS OF SINGLE CELLS INFORMED BY PHYSIOLOGY SUCH AS THE WORK PRESENTED EARLIER, SCALE THEM WHERE WE CAN TO THE LEVEL OF NEURAL SYSTEMS AND MAKE PREDICTIONS ABOUT BEHAVIOR IN SPECIFIC CASES. IN TURN, WHICH WOULD REALLY IS IMPORTANT ABOUT THIS FRAMEWORK, WE CAN PUT IN PREDICTIONS INTO SUCH A MODEL AND TEST THEM CAUSALLY IN HUMANS USING PHARMACOLOGICAL MANIPULATIONS AND COMPARE THEM TO CLINICAL EFFECTS IN PATIENTS. THIS BUSY PICTURE REALLY MAKES A SIMPLE 90 NOT ALL COMPUTATIONAL STUDIES ARE USEFUL FOR ALL LEVELS OF EXPERIMENTAL ANALYSIS. THEY HAVE TO BE ALIGNED. SO, ON THE LEFT YOU SEE EXPERIMENTAL LEVELS OF ANALYSIS GOING FROM CIRCUIT TO BEHAVIOR AND ON THE RIGHT YOU SEE COMPUTATIONAL MODELS BUILT TO INFORM EACH LEVEL. IDEALLY WE STRADDLE THIS BUT IT'S DIFFICULT TO DO BUT WE ARE BASICALLY WHAT WE ARE IN THE BUSINESS OF DOING TRYING TO ACCOMPLISH IS THE MODELS THAT STRADDLE THIS HIERARCHY. I'M GOING TO TALK ABOUT LOCAL SYNAPTIC SHIRTS WE TRY TO MODEL AND ULTIMATELY -- CIRCUITS -- THE AND WORK THAT IS MOST UNDER DEVELOPED, THE LARGE-SCALE NETWORK MODELING TO INFORM BIOMARKER DEVELOPMENT. SO WE BUILD MODELS AT EACH LEVEL TO PREDICT BEHAVIOR, TO PREDICT NEUROIMAGING EFFECTS AT THE NEUROSYSTEMS AND ULTIMATELY LARGE-SCALE FUNCTIONAL CONNECTOMICS THAT CAN REVEAL BIOMARKERS. SO LET'S GET STARTED. SO FOR THOSE OF YOU WHO DON'T THINK ABOUT PHARMACOLOGICAL MODELS AND SCHIZOPHRENIA, THE WAY WE MODEL DISEASE IN HEALTHY HUMANS, WE ADMINISTER ANESTHETIC BUT IN LOW DOSE INDICATE MEAN, NMDA RECEPTOR ANTAGONIST, INDUCES MAJOR SYMPTOMS. IT MAKES PEOPLE PSYCHOTIC. BUT IT IS REVERSIBLE DUE SHORT HALF-LIFE AND IT'S SAFE BUT IT'S CAUSAL MANIPULATION ON TO THE CIRCUIT. SO, JUST BRIEFLY OUTLINE THE MECHANISM HERE. SO THIS IS A PICTURE OF LEFT LATERAL PFC ZOOMING IN ON A SLICE CROSS SECTIONAL XO AND LOOKING AT A CANONICAL MICROCIRCUIT WE HEARD IN THE FIRST TALK AND THESE ARE THE DIFFERENT CELL TYPES IN SUCH A CANONICAL CIRCUIT, TWO BROAD CLASSES THAT WE HEARD ABOUT WHICH IS INTERNEURONS AND PARAMEATAL LLS CODED IN RED AND BLUE RESPECTIVELY. AND WHAT YOU HAVE IN EACH OF THESE CELL CLASSES IS NMDA RECEPTOR WHICH IS LIKE A LOGICAL GATE. IT'S A COEBBS DENSE DETECTOR DETECTING GLUTAMATE BINDING -- COINCIDENCE DETECTOR. WHAT WE THINK IS THAT THERE IS A DIVERSITY OF NMDA RECEPTOR SUBTYPES ACROSS DIFFERENT INTERNEURON TYPES SUCH THAT KEMAMINE, WHEN GIVEN AT A LOW DOSE, SLIGHTLY PREFERENTIALLY ANTAGONIZES CELLS THAT HAVE NR2C TYPE RECEPTOR AND THOSE HAPPEN TO BE ON INHIBITORY CELLS IN THE ADULT MAMMAL. AND THAT IF YOU'RE FOLLOWING THE LOGIC, SHOULD DISINHIBIT THE CORTICAL MICROCIRCUIT, ELEVATING THE FIRING RATE IN-VIVO AND THAT IS WHAT WE ARE PREDICTING AND THAT IS WHAT WE ARE BUILDING INTO OUR MODEL. SO I'M NOT GOING TO WALK THROUGH THE LOGISTICS OF THE DATA ACQUISITION BUT SUFFICE TO SAY FOR A PERSON TO UNDERGO THIS STUDY THEY HAVE TO UNDERGO 20 HOURS WORTH OF SCREENING AND SAFETY ASSESSMENTS FOLLOWING WHICH THEY COMPLETE RIGOROUS SCANNING DATA ABOUT FIVE HOURS IN THE MORNING, THE DATA IS THEN CRUNCHED ON OUR SUPER COMPUTER AND I'M PROUD TO SAY WE HARMONIZED THE NIH EARLY INDEPENDNCE AWARD ACQUISITION WITH A HUMAN CONNECT OHM PROJECT FULLY AT THE OUTSET. AND ALSO THE ANALYTICS OF THIS ARE RELATIVELY COMPLICATED BOTH IN TERMS OF NEUROINFORMATICS AND COMPUTATIONAL PRINCIPLES WE USED AND THE ADVANTAGES WE GET FROM MULTIMODAL INTEGRATION THIS STUDY IS DOING. SO I'M HAPPY TO WALK THROUGH THIS IN MORE DETAIL WITH PEOPLE OFF LINE IF THEY HAVE QUESTIONS. THAT SAID, HOW DO WE TRANSLATE A COGNITIVE FRAMEWORK TO COMPUTATIONAL MODEL? SO THIS IS A TASK, AN EXAMPLE, THAT A MONKEY CAN DO AND I DON'T MEAN THAT INSULTINGLY. THEY PERFORM THIS TASK WHERE THEY REMEMBER A LOCATION IN SPACE FOR A PERIOD OF TIME AND THEN THEY LATER REPORT WHERE THEY SAW IT. DURING PERIOD, CELLS IN THE PRINCIPLE OF THE LATERAL PFC BEGIN TO FIRE UNABATED AND THIS IS ESSENTIALLY A COMPUTATIONAL PRINCIPLE OF HIRE COGNITION THAT FOLKS LIKE PATRICIA AND OTHERS HAVE CHARACTERIZED AT YALE. IMPORTANTLY, THE COMPUTATIONAL PRINCIPLES OF THIS FRAMEWORK HAVE BEEN WORKED OUT BY FOLKS BEFORE US AND AGAIN, THE TWO PRINCIPLE CLASSES OF CELLS, THE EXCITATORY AND INHIBITORY CELLS THAT FORM A CANONICAL MICROCIRCUIT, SHAPE THE BALANCE AND DUE TO THE SLOW REVERBATION OF THE NMDA RECEPTOR, THEY CAN STAY ON EVEN IN THE ABSENCE OF A STIMULUS, PARTICULARLY THE PREFRONTAL CORTEX ONES AND YOU'RE SEEING IN THE GRAPH ON THE X AXIS IS TIME AND THE Y AXE SIS SPACE. AND THERE IS A CURRENT INJECTED IN SOME NEURONS AND THEY CONTINUE TO FIRE. NOW WHAT HAPPENS WHEN WE DISINHIBIT THE CIRCUIT WHEN WE PUT IN MANIPULATION WHEN WE PREDICT KEMAMINE IS ACCOMPLISHING? THESE EQUATIONS HIGHLIGHT THE CONDUCTANTS OF THE NMDA RECEPTOR AND ALL WE ARE DOING IS DIALING DOWN THE NMDA CONDUCTANTS ON I CELLS FROM E CELLS SUCH WE DISINHIBIT THE CIRCUIT. IN TURN THIS BROADENS THE MEMORY. SO IN OTHER WORDS, THE CANONICAL MICROCIRCUIT IS NOT AS PRECISE. IT CONTINUES TO FIRE UNA BAITED BUT IT SIMPLY WIDER REPRESENTING MORE OF THE INFORMATION. NOW THAT IS PROBLEMATIC BECAUSE IF YOU ARE NOT PRECISE AND IF YOU'RE DISTANTED, WHICH WE CAN DO HERE BY SHOWING A LOCATION, IN A DIFFERENT LOCATION, WE CAN ESSENTIALLY BIAS THE MODEL, ESPECIALLY IN THIS INHIBITED RES SCREAM TO REPORT MORE WITH HIGH LIKELIHOOD THAT IT SAW SOMETHING WHEN IT DIDN'T. SO THAT IS THE KEY PREDICTION. THAT IF WE WERE TO GIVE KEMAMINE TO HEALTHY PEOPLE, THAT IN THE CONTEXT OF A PARADIGM LIKE THIS THEY WOULD LIKELY TO BE FALSE ALARM AND SAY THEY SAW THINGS WHEN THEY IN FACT DIDN'T AND THIS CAN BE QUANTIFIED USING A DISTRACTIBILITY FRAMEWORK WINDOW WE USE. SO THE TASK THAT WE TESTED THIS WITH IS A VERY SIMPLE DELAYED WORKING MEMORY TASK. PEOPLE REMEMBER LOCATIONS IN SPACE OVER TIME AND THEN LATER THEY SAY IF THEY SAW SOMETHING, IN A GIVEN LOCATION OR THEY DIDN'T, NOW DURING THIS ACQUISITION, PEOPLE GET EITHER SALINE OR KEMAMINE WHILE PERFORMING THIS TASK. THEY DO WORSE ON KEMAMINE. SO THIS GRAPH SHOES YOU THERE IS A DROP IN WORKING MEMORY WHILE PEOPLE ARE PERFORMING THIS. YOU CAN SEE THE ORANGE BARS SHOWS ELEVATED FALSE ALARM RATE WHEN THE NON TARGET PROBES ARE NEAR THE LOCATION OF THIS INHIBITION. NOW, THE IMPORTANT BIT OF THIS KIND OF COMPUTATIONAL PSYCHIATRY FRAMEWORK IS THAT WE CAN PARAMETERIZE THEM ALL. WE CANNOT RUN 50 DIFFERENT EXPERIMENTS TO TEST THE RANGES AND DOSES IN HEALTHY HUMANS BECAUSE IT'S NOT SAFE AND COST PROHIBITIVE BUT IN THE MODEL, WE CAN GET A REGIME THAT SUGGESTS THE ENTIRE ACTIVITY COLLAPSES IF WE ANTAGONIZE EXCITATORY CELLS F WE ANTAGONIZE INHIBITORY CELLS WE GET DISINHIBITED REGIME OR A BALANCE. SO WE HAVE A COMPETING FRAMEWORK WE CAN TEST USING OUR EXPERIMENT AND IN OUR HANDS, WE SEE EVIDENCE OF THIS INHIBITION. SO WE CAN ZOOM THIS NOW TO DISTRICTED CIRCUITS TO MAKE PRESCRIPTIONS ABOUT BRAIN ACTIVITY. WHEN WE TESTED WHAT KETAMINE DOES ON BRAIN ACTIVITY PATTERNS DURING THIS PARTICULAR TASK, WE FOUND THAT ESSENTIALLY BIDIRECTION ALLEY MODULATES TASK EVOKED AND TASK-SUPPRESSED SIGNALS. SO THIS IS A CANONICAL TRACE OUT OF A LATERAL PFC AREA AND MORE POSTERIOR AREA AND THE RED TRACE THAT THE ANTAGONISM SQUISH THIS IS BIDIRECTIONALLY AND THAT IS IMPORTANT FOR A NUMBER OF REASONS BUT WHAT IS THE MECHANISM? HOW DO WE BEGIN TO THINK ABOUT THIS HAPPENING AT THE NEURAL SYSTEM LEVEL? SO WE CAN TAKE THE SAME MODEL AND EXTEND IT TO HAVE AN ADDITIONAL MODULE WHICH WE WOULD EN VIEW WITH A HIGH BASELINE ACTIVITY AND MAKE A DEACTIVATED TASK ONSET BY MAKING IT RECIPROCALLY INHIBITORY WITH THE TASK POSITIVE MODULE AND WE CAN GET A FIRING RATE WHICH WE CAN INVOLVE WITH THE BALLOON MODEL TO GET THE BOLT SIGNAL AND YOU SEE THESE SMOOTH TRACE SYSTEM CAPTURE BIDIRECTIONAL ATTENUATION FOLLOWING KETAMINE ADMINISTRATION. WE CAN PARAMETERIZE IT. THIS IS A PARAMETER SHAPE HIGHLIGHTS TWO DEPENDENT MEASURESSED PLOTTED AND YOU CAN SEE AGAIN WE SEE A DECREASE IN BOTH MODULE FIRING TRACES IF WE ANTAGONIZE E TO E CONNECTIONS IF WE ANTAGONIZE E TO I CONNECTIONS, THE MODULES FLIP, WHICH INHIBITION OR THEY ARE BALANCED AGAIN PROVIDING A STRONG FRAMEWORK. ONE COMPETING FRAMEWORK WE CAN TEST, WHICH IN OUR HANDS THE RESULTS ARE MORE CONSISTENT WITH THIS INHIBITION. SO AGAIN, WE CAN START FROM A COMPUTATIONAL MODEL TO GENERATE TESTABLE PREDICTIONS PARAMETERIZE IT TO GENERATE TWO COMPETING FRAMEWORKS, TEST THEM EXPERIMENTALLY AND THEN FINALLY REANALYZE OUR CLINICAL FRAMEWORK IN MIND AND GET EVIDENCE THAT THERE IS A QUIT EFFECT THROUGHOUT. AND WE SUMMARIZED THAT THIS SUGGESTS THIS TYPE OF DISTURBANCE THAT WE SEE ON KETAMINE MIGHT BE PRESENT ACROSS THE NEUROPSYCHIATRIC SPECTRUM AND MIGHT NOT BE SCHIZOPHRENIA SPECIFIC. SO TO BRIEFLY SUMMARIZE BEFORE I GET TO THE LAST BIT, HOPEFULLY I'M SHOWING YOU THAT WE CAN USE A MICROCIRCUIT FRAMEWORK TO START TO GENERATE BOTH PRICE BEHAVIORAL PREDICTIONS AS WELL AS SCALE IT TO THE LEVEL OF NEURAL SYSTEM TO GENERATE EFFECTS IN THE BRAIN THAT WE CAN MEASURE WITH THE BOLT SIGNAL AND EVEN SMALL AMOUNTS OF THIS INHIBITION CAN PRODUCE PROFOUND EFFECTS ON BEHAVIORAL. NOW FINALLY, IF I DON'T RUN OUT OF TIME, THIS LAST BIT IS WHAT I'M MOST EXCITED ABOUT BECAUSE IT IS AN ELEMENT THAT THE NIH STRATEGIC PLAN ARTICULATES TO MAP THE CONNECT OHMS FOR MENTAL ILLNESS. THE TOOLS WE ARE USING TO STUDY LARGE-SCALE NEUROSYSTEMS IN HUMANS HAVE MATURED ENOUGH THAT WE CAN USE THEM TO PROBE ALTERATIONS IN PSYCHIATRIC DISEASE. SO, THE MODALITY THAT I'LL TALK ABOUT IS RESTING STATE FUNCTIONAL MRI. SOME MAY HAVE HEARD ABOUT IT. THOSE WHO VENT, THE IDEA IS IF YOU PUT A PERSON INSIDE I A MRI MACHINE AND ASK THEM TO DO NOTHING IN PARTICULAR OTHER THAN LAY THERE, YOU CAN RECORD THE IF YOU CANNATION OF BOLD SIGNAL OVER TIME AND IT TURNS OUT THAT DISTRIBUTED LARGE-SCALE SYSTEMS SHOW COHERENT SPACIAL TEMPORAL FLUCTUATIONS. AND NOBODY IS DOING ANYTHING IN PARTICULAR. THIS IS SORT OF A SCAFFOLD OF THE UNDERLYING FUNCTIONAL SYSTEM WE SEE ACROSS MAMMALIAN SPECIES BUT IN HUMANS IN PARTICULAR. THE FRAMEWORK IS FLEXIBLE. WE CAN USE MANY DIFFERENT PHYSICAL APPROACHES TO TRY TO UNDERSTAND WHAT IS DRIVING THESE PATTERNS. WE CAN CAN BE FULLY DATA DRIVEN AND FASTER AND CHEAPER APPROACH TO GET AN ASSAY OF THESE NEURAL SYSTEMS. SO IS IT USEFUL? HAVE WE MADE PROGRESS USING THIS TOOL? WE STARTED OFF IN 1995 WITH OBSERVATION THAT LEFT AND RIGHT MOTOR CORTEX EXHIBIT COHERENCE AT REST IGNORED FOR 10 YEARS AND IF WE FAST FORWARD TO 2011, WORK FROM RANDY BUCKNER CHARACTERIZED LARGE-SCALE NEUROSYSTEMS ACROSS THOUSANDS OF PEOPLE AND IT'S FAIR TO SAY WE CAN SCAN ANYBODY IN THIS ROOM FOR 60 MINUTES AND CAPTURE ALL OF THESE LARGE-SCALE NEURAL NETWORKS RELIABLY. IF WE LOOK IN 2016 NOW, THE HUMAN CONNECT OHM PROJECT UNDER BUDGET AND UNDER DEADLINE HAS PRODUCED WHAT IS THE COMPREHENSIVE HUMAN CORTEX ACROSS VARIOUS MODALITIES. IT'S NOT DEFINITIVE SO THERE WILL BE ITERATIONS BUT IT IS COMPREHENSIVE. SO I HOPE THAT IT GIVES US A RESOUNDING YES THIS STOOL USEFUL TO BE APPLIED TO PSYCHIATRIC ILLNESS. SO THE APPROACH, SIMPLE APPROACH THAT I WANT TO CONCLUDE WITH IS THE IDEA OF SEED-BASED CONNECTIVITY. YOU HAVE TO HAVE A HYPOTHESIS. HAVE TO KNOW WHERE TO START. SO WE START FROM THE THALAMUS WHICH IS A FASCINATING STRUCTURE LIKE A SHRINK WRAPPED VERSION OF THE NEOCORTEX THAT FORMS PARALLEL AND SEGREGATED PATHWAYS THROUGHOUT THE CENTRAL INFORMATION SYSTEM. IT'S FUNCTIONALLY CONNECTED TO THE ENTIRE CORTEX AND READILY DEFINED USING IMAGING TOOLS SO IT HAS IMPORTANT FEATURES IN THAT WAY. SO THIS PICTURE SHOWS YOU A CONTRAST BETWEEN 90 PATIENTS SCHIZOPHRENIA AND 90 HEALTHY CONTROLS IN THEIR PATTERN OF MISCOMMUNICATION, EVERYTHING THAT IS WARM IS ELEVATED IN SCHIZOPHRENIA. EVERYTHING THAT IS BLUE IS REDUCED IN TERMS OF COMMUNICATION WITH THE THALAMUS. SO YOU CAN SEE FUNCTIONAL SEGREGATION EMERGE WHERE THE WARM AREAS ARE ENTERED ON SENSORY MOTOR AND THE BLUE ONES ARE MORE ASSOCIATIVE AND THE TWO PATTERNS ARE HIGHLY RELATED AND PATIENTS WITH SCHIZOPHRENIA ARE SHOWING RED IN THIS PLOT SITTING IN THE LOWER RIGHT QUADRANT. SO THERE IS A BLURRING OF INFORMATION FLOW THROUGH THE LARGE-SCALE SYSTEMS IN PEOPLE WHO ARE ON THE SCHIZOPHRENIA SPECTRUM. AND BIPOLAR PATIENTS ARE SHOWN IN BLUE AND THEY ARE NOT SERIOUSLY SHIFTED. NOW, THE QUESTION IS, WE SEE THIS IN KNOW CROIC PATIENTS BUT DOES IT EMERGE IN THE PRODROME, IN THE EARLY STAGES OF CLINICAL RISK PRIOR TO FULL DIAGNOSIS? SO WE DID THIS IN COLLABORATION WITH THAI AT YALE AND WE REPEATED THIS ANALYSIS ONLY TO FIND THE SAME EFFECT IN PEOPLE WHO ARE CLINICALLY AT RISK THAT ARE NOT SICK YET. AND ESSENTIAL NETWO AREAS THAT I'LL HIGHLIGHT, THE MAIN POINT IS THE RED BARS AND THE RED DISTRIBUTION PLOTS. THEY ARE SHIFTED MORE SERIOUSLY AND THOSE ARE THE PEOPLE WHO ARE AT CLINICAL RISK BUT GO ON TO DEVELOP THE FULL ILLNESS ONE YEAR LATER. THEY ARE SCANNED AT BASELINE AND THEN THEY GET SICK LATER AND THEY ARE PARTICULARLY ALTER 8ING IN THIS PATTERN. AGAIN, THESE ARE NOT INDEPENDENT SOURCES OF DISTURBANCE. THE HYPERAND HYPOCONNECTIVITY SCALE. WE CAN CAPTURE THE SAME EFFECT USING KETAMINE. SO WHEN WE ADMINISTER KETAMINE TO HEALTHY INDIVIDUALS, THIS IS WORK BY MY STUDENT CHARLIE, WE CAN BASICALLY SEE THE SAME PATTERNS. SO BIDIRECTION ALPER TERBATION AND IT'S LINEARLY RELATED CROONS PILOT. -- WE WENT TO A RARE GENETIC SYNDROME WHERE WE KNOW THE REASON FOR THIS CONNECTIVITY 22Q11 DELETION SYNDROME AND SO HOW DO WE MODEL THIS? AND I HAVE ABOUT TWO MINUTES TO TELL YOU ABOUT THIS. SO, WE WENT INTO THE STORY BY TRYING TO CHARACTERIZE THE VARIANCE, VARIABILITY IN THESE LARGE-SCALE NEURAL SYSTEMS IN SCHIZOPHRENIA FOR METHODOLOGICAL REASONS BUT MY GRADE STUDENT, REALLY TOOK THIS STORY FORWARD AND -- GRADUATE STUDENT -- AND CHARACTERIZED THE FUNCTIONAL PATTERNS OF A THING WE TYPICALLY DISCARD IN IMAGING. WE JUST REMOVE IT, THE GLOBAL SIGNAL. SO WE AVERAGE SIGNAL EVERYWHERE AND REGRESS IT OUT BECAUSE WE THINK IT CAPTURES ARTIFACT VERY WELL. NOW THIS APPEARS TO BE WRONG. SO SHE WAS ABLE TO FIND ELEVATEED VARIANCE REPLICATED THAT ELEVATED VARIANCE AND DIDN'T FIND IT IN BIPOLAR PATIENTS AND IT WAS CONSISTENT. THE VARIABILITY OF THIS LARGE SCALE BACKGROUND FLUCTUATION SEEMS TO BE ELEVATED IN SCHIZOPHRENIA AND TYPICALLY REMOVE IT. SO, SHE WANTED TO MODEL THIS. SHE WANTED TO COMPUTATIONALLY SCALE THE LIKE ROW CIRCUIT FRAMEWORK TO THE LEVEL OF NEURSYSTEM TO UNDERSTAND THIS. SO WE HAVE OUR SAME MODEL THAT IS NOW FOLDED INTO ONE FUNCTIONAL UNIT ALTERING THE EI BALANCE AND EXPANDED TO A NUMBER OF DIFFERENT REGIONS TO CAPTURE INTERACTION SCHEMES CROSS-FUNCTIONAL NETWORKS. AND THE IMPORTANT BIT OF THIS MODEL ISED THAT BY PHYSICALLY GROUNDED, IT'S CONSTRAINED BASED ON CONNECTIVITY FROM HEALTHY HUMANS AND IN TURN, IT'S FITTED TO THE FLUCTUATIONS OVER TIME IN THE SAME MATRIX THAT IS SIMULATED IN THE MODEL AS WELL AS HUMANS. SO WHAT SHE FOUND IS WHEN SHE TWEAKED TWO KEY PARAMETERS THAT ARE FUNCTIONALLY MEANINGFUL, SHE FOUND ELEVATED VARIABILITY IN THE MODEL IN SILICO THAT BEGINS TO CAPTURE WHAT WE SEE IN PATIENTS WITH SKITS 49IA AND IT DOESN'T BREATHE OR MOVE OR HAVE ANY ART FACT. -- SCHIZOPHRENIA. WHEN SHE WENT DEEPER, SHE FOUND THAT THERE IS AN ALTERATION SPECIFICALLY IN ASSOCIATION. SO HIGHER ORDER NETWORKS, IN TERMS OF CONNECTED ACTIVITY AND VARIANCE THAT WAS NOT THERE IN SENSORY NETWORKS. WHICH SAYS THE MODEL IS WRONG. WE CAN'T USE A HOMOGENEOUS MODEL TO EXPLAIN THIS BECAUSE IT'S NOT HOMOGENEOUS. THEY ARE PREFERENTIAL EFFECTS. THESE ARE LINEAR RELATED. SO WE WENT BACK TO PRE-CLINICAL WORK TO IDENTIFY HIERARCHY OF INTRINSIC TIME SCALES ACROSS CORTEX THAT CHANGED SUGGESTING THAT A SO ISATIVE CORTEX REGIONS HAVE A SLOWER TEMPORAL DECAY, DIFFERENT DYNAMICS THAN THOSE IN SENSORY REGIONS AND WE VIEWED THE MODEEL WITH THIS. SO WE TOOK THE SIMPLE MODEL I SHOWED YOU. WE SPLIT IT INTO TWO HIERARCHAL SYSTEMS THAT ARE RECIPROCALLY CONNECTED USING DIFFUSION WEIGHTED ANATOMICAL CONNECT ACTIVITY AND GAVE THEM DIFFERENT REOCCURRENCE PATTERNS. IN THE CASE WHERE THE REOCCURRENCE PATTERNS ARE CONITANT, IT DOESN'T EXPLAIN DIFFERENCES BETWEEN MODULES. THEY ARE THE SAME. BUT WHEN WE MAKE ONE MODEL MORE RECURRENT, PERHAPS THE GEOMETRY OF THESE CORTICAL MICROCIRCUITS, YOU SEE DYSFUNCTIONAL DIFFERENTIATION SUCH THAT SO ISATIVE CORTEX IS MORE NORMAL TO THE SIM INTRINSIC UNDERLYING PERTIVATION WHICH IS WHAT WE OBSERVED IN PATIENTS. SO TO SUMMARIZE, HOPEFULLY I SHOWN YOU HOW ALTERATIONS IN LOCALLY GLOBAL COUPLING OF THESE MODELS CAN CAUSE ELEVATED VARIANCE AND ELEVATED CONNECT ACTIVITY THEY SIMILAR TO CLINICAL AND PHARMACOLOGICAL EFFECTS ADDING PHYSICALLY-BASED FEATURES SUCH AS HIERARCHY INTO THIS MODEL IS VITAL TO BETTER EXPLAIN THE EFFECTS AND ULTIMATELY HOPEFULLY EXTENDS THE MICROCIRCUIT BREAKER WORK TO ENROLL SYSTEM LEVEL TO GIVE US AN ANCHOR ON THESE BIOMARKERS WE ARE TRYING TO KIDNAP. HOPEFULLY THIS COALESCED -- WHERE WE HAVE SHOWN EVIDENCE THE MODEL AND PHARMACOLOGY CAN INTERPLAY AT EACH LEVEL. THANK YOU FOR YOUR ATTENTION AND I'LL STOP THERE. [ APPLAUSE ] >> I'M CURIOUS WHAT HAPPENS IN PEOPLE THAT START TAKING MEDICATION TO ALLEVIATE SOME OF THE SYMPTOMS OF SCHIZOPHRENIA OR THE KIND OF -- MAKE THEM ACT MORE NORMAL. WHAT HAPPENS WHEN THE LEVEL OF BRAIN CONNECTIVITY? >> WHEN WE ADMINISTER MEDICATION? >> PEOPLE THAT ARE ON DRUGS TO TREAT. >> SO, BASICALLY, MOST TREATMENT CLASS SYSTEM THAT WE HAVE RIGHT NOW ARE DOPAMINERGIC, TICKLY D2 ANTAGONISTS THAT ACT AT THE LEVEL OF THE STREET UM SO YOU SEE NORMALIZATION OF THESE PATTERNS BUT THEY ARE NEVER GLUTAMATERGIC IN ORANGE IN. SO IF WE BELIEVE THOSE DEFICITS ARE UPSTREAM OF THE DOPAMINERGIC ONES, THEY ARE NEVER REALLY FIXING THE UNDERLYING PROBLEM. WHAT WE DO SEE AND THERE IS NOT A LOT OF DATA TO LOOK AT PRE-POST CLINICAL TRIAL TREATMENT EFFECTS IN SCHIZOPHRENIA. WE SEE A NORMALIZATION OF THIS HYPERCONNECTIVITY THAT I SHOWED YOU SO THAT DOES ATTENUATE IN EARLY CORE SCHIZOPHRENIA. WE HAVEN'T LOOKED AT THE CORTICAL PATTERN. BUT, THE OTHER THING THAT I CAN SAY IS THAT THERE ARE FEATURES OF THE ELEVATED VARIANCE THAT SEEM TO PERSIST THAT WE CAN'T FULLY ALLEVIATE OR ATTENUATE. SO ABOUT IT'S A GREAT OPEN QUESTION. >> A LEADING CANDIDATE FOR THE SUPPLY OF A COAGONIST, AND ALSO FOR COUPLING BETWEEN NEURONAL ACTIVITY AND BLOOD FLOW, IS IS IT INTERVENING CELLS CALLED THE ASTROCYTE AND INDICATE MEAN EFFECTS ASTROCYTES. HOW DO YOU INTERPRET THAT WITH YOUR MODEL WHICH SEEMS TO ASSUME THAT KETAMINE ACTS ENTIRELY ON THE NEURONS AND NOT ON AN INTERVENING STEP? >> SO THAT'S A GREAT QUESTION. SO KETAMINE IS BY NO MEANS A CLEAR DRUG. SO I DON'T WANT YOU TO WALK AWAY WITH THE IMPRESSION THERE IS SPECIFICITY IN THE EFFECTS. IT HAS OTHER TARGETS OF ACTION AND IT ACTS IN THE CHIINATORY CEPTOR AND CERTAINLY TO THE DOPAMINERGIC SYSTEM DUE TO INDIRECT EFFECTS. THE WAY I THINK ABOUT THIS PROBLEM IS, WHAT IS THE NET EFFECT? WHAT IS IT ITS PRINCIPLE NET MECHANISM OF ACTION? IF YOU TAKE ALL OF THESE OTHER OFF TARGET EFFECTS AND IT SEEMS AS IF IN OUR HANDS, THE ANTAG - AUTO DISRUPTION OF EI BALANCE IS BOTH NECESSARY AND SUFFICIENT TO GET THE BARREL AND PHARMACOLOGICAL IMAGING EFFECTS TO EMERGE -- IF WE START TO INCORPORATE FURTHER DIVERSITY WHICH IS A CRITICAL PIECE HERE, AS WELL AS ASTROCYTES AND AGREEMENT CELLS, WE MIGHT GET A FINER PICTURE OF HOW THIS WORKS BUT RIGHT NOW YOU'RE RIGHT. THIS IS SIMPLISTIC -- ONE INTERNEURON BUT TWO CELL TYPE INTERACTION MECHANISMS THAT EVEN IN THE SIMPLICITY SEEMS TO BE ABLE TO CAPTURE SOME OF THE CORE EFFECTS. >> THANK YOU. >> SO HOW DO YOU CONSTRAIN YOUR PARAMETERS IN YOUR MODEL TO GET USEFUL PREDICTION AND NOT OVER-TWEAKING AND HOW DO YOU DEAL WITH VARIABILITY THAT YOU SEE IN YOUR DATA THAT IS NOT NORMALLY DISTRIBUTED? >> WELL, LOOSE TAKE THE STATISTICAL QUESTION FIRST. -- LET'S TAKE -- AND OBVIOUSLY WE USE THE STANDARD CORRECTIONS FOR ABNORMALITIES TRY TO BLOT OUR DATA WHEN POSSIBLE AND ADJUST OUR INFERENCE IF NOR MALTY IS SHARPLY VIOLATED. BACK TO THE FIRST QUESTION, ONE KEY PARAMETER IS NOT HAVING MANY FREE PARAMETERS, HAVING NO MORE THAN FREE PARAMETERS IN A MODEL SO THE KEY PARAMETER WE ARE RETURNING IS THE NMDA CONDUCTANTS ON EACH I CELL. EVERYTHING ELSE IS PRETTY MUCHELED CONSTANT IN THE MODEL I SHOWED YOU. -- WHERE YOU START TO INCORPORATE SYNAPTIC CONNECTION PATTERNS TO LARGE-SCALE AREAS, THE NUMBER OF PARAMETERS GROWS AND THIS BECOMES MORE AND MORE OF A CONCERN BUT RIGHT NOW WE ARE BASICALLY DON'T INTRODUCE A PARAMETER UNTIL WE HAVE VERY HIGH LEVEL OF COMPETENCE THAT IT HAS BIOPHYSICAL ACTIVITY. FOR INSTANCERS THE MORPHOLOGY OF MART NOTCHY CELLS IS SOMETHING WE WOULD PUT INTO THE MODEL. >> CAN YOU DO SOMETHING LIKE TAKE OUT YOUR DATA TO CONSTRAIN THE PARAMETER AND THEN DO A PREDICTION FOR ANOTHER SET OF DATA TO CROSS VALIDATION AND REALLY COMPARE THAT TO MODEL COMPLEXITY AND THEN ALSO THE MODEL IS BASED ON MEANS. OR IF YOU DON'T REALLY TAKE MODELING FRAMEWORK VARIABILITY INTO CONSIDERATION IF I UNDERSTAND CORRECTLY. >> A GREAT QUESTION. SO IN THE FIRST BIT THAT I SHOWED YOU, YOU'RE CORRECT. WE TAKE THE MODEL POPULATION VECTOR ACTIVITY AT THE READ OUT STAGE AND GET A MEAN VALUE AND THEN WE GET THE GROUP MEANS FROM OUR SAMPLE AND WE QUANTITATIVELY FIT THEM. BUT WE DON'T FIT THEM AT THE INDIVIDUAL SUBJECTED LEVELS. SO ONE THING WE ARE UP TO AND IN FACT, WE HAVE A GRANT YOU SHOULD REVIEW ON THIS TOPIC, IS TO TRY TO FIGURE OUT WHETHER WE CAN MEASURE INDIVIDUAL SUBJECT NEUROIMAGING DATA OVER TIME AND TAKE THE MODEL CONSTRAIN IT BASED ON DIFFUSION IMAGING PATTERN FROM THAT PARTICULAR PERSON, SO INDIVIDUAL-SPECIFIC CONNECTOMICS AND THEN ACHIEVE FIT IN HIERARCHAL WAY. BUT FIRST AT THE SINGLE-SUBJECT LEVEL AND THEN DO THE GROUP FITTING TO INCORPORATE THE SOURCE OF VARIABILITY YOU ARTICULATE, IN DIFFERENT PEOPLE MIGHT HAVE DIFFERENT PERTERBATIONS ACROSS THESE NETWORKS. >> THANK YOU SO MUCH. THANK YOU TO ALL OF OUR SPEAKERS THIS MORNING. [ APPLAUSE ] WE ARE GOING TO TAKE A 5 MINUTE BREAK. 15. EXCUSE ME. A 15 MINUTE BREAK. I WANT TO ENCOURAGE YOU TO CONTINUE THE DISCUSSION WITH THE SPEAKERS WHO ARE HERE IN THE FRONT AND WE'LL SEE YOU BACK IN 15 MINUTES. >>> I'D LIKE TO INTRODUCE OUR FIRST SPEAKER OF THE SECOND SESSION WHO WILL TALK ABOUT SYNTHETIC BIOLOGY PLATFORMS FOR NATURAL PRODUCT BIOSYNTHESIS AND DISCOVERY. CHRISTINA? >> SO, I'M GOING TO SHARE WITH YOU SOME OF THE WORK THAT WE HAVE DONE THAT WAS FUNDED THROUGH A PIONEER AWARD THAT WE RECEIVED SEVERAL YEARS AGO NOW. I'LL START OFF WITH JUST A LITTLE BIT OF MOTIVATION FOR THE WORK. SO, ABOUT OVER HALF OF THE MEDICINE THAT IS CURRENTLY APPROVED ARE COMPOUNDS THAT COME FROM NATURE AND A LITTLE BIT OVER HALF OF THOSE COME FROM THE PLANT KINGDOM. AND A LOT OF MY TALK TODAY OR SOME OF MY TALK WILL FOCUS ON ONE IN PARTICULAR TYPE OF MEDICINE THAT DERIVED FROM OPIUM POPPY AND MANY ARE FAMILIAR WITH OPIOIDS USED THAT ARE EFFECTIVE PAIN MEDICINE. BUT IF YOU LOOK ACROSS THE PLANT KIM DOMESTIC, YOU CAN FIND -- KINGDOM -- ESSENTIAL MEDICINE DERIVED OR INSPIRED FROM COMPOUNDS IN PLANTS THAT ARE USED TO TREAT CANCER, COMPOUNDS USED TO TREAT INFECTIOUS DISEASE AND ALSO COMPOUNDS USED TO TREAT HEART DISEASE AND HYPERTENSION. AND SO THERE IS A BROAD DIVERSITY OF DIFFERENT PHARMACOLOGICAL ACTIVITIES THAT WE GET FROM THE PLANT KINGDOM. AND IF YOU THINK ABOUT IT, WHEN WE ARE USING PLANTS TO SOURCE OUR MEDICINE, THERE ARE A LOT OF CHALLENGES AND I'LL STEP THROUGH SOME OF THE EXAMPLES AS WE THINK ABOUT JUST THE SUPPLY CHAIN AROUND HOW WE SOURCE OPIOIDS. SO ALL OF OUR MEDICINAL OPIOIDS DERIVED FROM THE MOREOVER IN SCAFFOLD COME FROM OPIUM POPPY FARMING. SO THIS IS A PICTURE OF A OPIUM POPPY FARM IN TASMANIA WHICH SOURCES ABOUT HALF OF THE LEGAL OPIATES GLOBALLY. AND THIS IS A PICTURE OF THE POD OF THE PLANT WHERE THE ALKALOIDS ACCUMULATE AND THE PLANT HAS EVOLVED OVER MANY HUNDREDS OF THOUSANDS OF YEARS TO MAKE ALL DIFFERENT TYPES OF OPIATES. SO YOU GO THROUGH A ANNUAL HARVEST WHERE YOU EXTRACT THE ALKALOID COMPOUND THROUGH THE PLANT MATERIAL AND THEN BECAUSE THE PLANTS ARE GROWN PRIMARILY NOT IN THE COUNTRIES WHERE THEY ARE USED YOU CAN GO THE EXPERT AND IMPORT PROCESS AND THEN WITHIN THE COUNTRY OF USE, IT GOES THROUGH FURTHER CHEMICAL PROCESSING AND IN THIS CASE, THE PLANT ISI MAAING A COMPOUND THAT IS NOT VIEWED AS THE MOST VALUABLE MEDICINE AND SO MOST OF THE MORPHINE THAT WE OBTAIN FROM THE OPIUM POPPY IS FURTHER PROCESSED AND CONVERTED THROUGH CHEMICAL SINT SIS ROUTES TO OTHER TYPES OF MEDICINE. SO IT CAN BE CONVERTED TO MEDICINAL OPIOIDS LIKE CODEINE, OXY, HYDROCODONE, ALSO ANTAGONISTS LIKE MALTEXONE AND NALOXONE USED PRIMARILY INCREASINGLY TO MAKE ABUSE FORMULATIONS OF OUR PAIN KILLERS AND PARTIAL AGONISTS WHICH ARE USED IN SOME CASES TO TREAT CHRONIC PAIN ESPECIALLY IN ADDICTS BECAUSE THEY HAVE A DECREASED SIDE EFFECTS, RESPIRATORY DEPRESSION AND DECREASED ADDICTION AND THERE ARE ALSO A SUITE OF OTHER MEDICINES DERIVED FROM THIS SCAFFOLD AS WELL. AND IF WE LOOK AT THE PLANTS, THE PLANT IS MAKING COMPOUNDS BEYOND JUST THE OPIATES. IT'S ALSO MAKING OTHER ALKALOIDS THAT HAVE DIFFERENT TYPES OF PHARMACOLOGICAL ACTIVITIES, COMPOUNDS USED TO TREAT CANCER, AS ONE EXAMPLE. AND SO, IF YOU JUST LOOK AT THE SUPPLY CHAIN AS A WHOLE AND WHEN WE USE OR GROW PLANTS FOR OUR MEDICINE, MANY CHALLENGES ARISE SO WE CAN CATEGORIZE INTO DIFFERENT AREAS. ONE IS OF COURSE YOU HAVE SUPPLY CHAIN RISKS AND VARIABILITY AND YIELDS DUE TO GROWING PLANTS IN THE OPEN FOR ABOUT A YEAR WHERE DIFFERENT THINGS CAN HAPPEN OVER TIME THAT CAN BE UNPREDICTABLE IN TERMS OF CLIMATE, CHANGES IN CLIMATE, DISEASE THAT COME THROUGH A PARTICULAR AREA, SORT OF LEFT WITH THE COMPOUND THAT THE PLANT HAS EVOLVED TO MAKE AND THE PLANTS ARE NOT MAKING COMPOUNDS FOR MEDICINE OR HUMANS SO AS WE SEEN WITH OUR PAINKILLERS, WE WIND UP HAVING COMPOUNDS THAT HAVE HORRIBLE SIDE EFFECTS WE HAVE TO DEAL WITH IN DIFFERENT WAYS. AND IN MANY CASES IT CAN BE DIFFICULT TO UPGRADE AND MODIFY THESE COMPOUNDS WE GET FROM NATURE. AND STILL TODAY, ALTHOUGH WE DON'T NECESSARILY THINK ABOUT THIS IN THE U.S. F WE LOOK GLOBALLY, MOST OF THE MEDICINES ARE UNAVAILABLE TO THOSE WHO NEED THEM MOST N CASE OF PAINKILLERS, THE WORLD HEALTH ORGANIZATION ESTIMATES 5 1/2 BILLION PEOPLE WHO DON'T HAVE SUFFICIENT ACCESS TO MEDICINES TO TREAT MODERATE TO SEVERE PAIN. AND JUST IF WE THINK BROADLY ABOUT THE ENTIRE CHEMICAL SPACE WE'D LIKE TO BE ABLE TO SEARCH IN TERMS OF DEVELOPING NEW MEDICINE, MOST OF THAT CHEMICAL SPACE WILL BE IN ACCESSIBLE THROUGH THIS ROUTE. AND SO WE WANTED TO, WHEN WE STARTED THIS PROJECT, THE PROGRAM ON MY LABORATORY, WE WERE STARTING WITH A FAIRLY SIMPLE QUESTION, COULD WE ACTUALLY BASICALLY CHANGE THE SUPPLY CHAIN AND THE MANUFACTURING PLATFORM FROM ONE THAT LOOKED LIKE THIS WHICH RELIED ON GROWING PLANTS OVER THE PERIOD OF A YEAR, TO ONE THAT LEVERAGED INSTEAD BIOTECHNOLOGY AND IN IN CASE, WHAT WE ARE LOOKING AT IS COULD WE INSTEAD TAKE A SIMPLE ORGANISM WE KNOW HOW TO GROW CHEAPLY AND RAPIDLY, LIKE BAKERS YEAST MODIFY THE GENETIC MATERIAL WITHIN THAT ORGANISM SO IT COULD FERMENT ON SUGAR BUT INSTEAD IN ADDITION TO MAKING COMPOUNDS THAT WE ARE VERY FAMILIAR WITH, LIKE TO BUILDUP VERY COMPLEX MEDICINAL COMPOUNDS. SO WHAT THIS REALLY REPRESENTSA A SHIFT IN TERMS OF OUR RELATIONSHIP WITHED AND HOW WE THINK ABOUT SOURCING AND DISCOVERY MEDICINE. SO AGAIN, WE ARE SHIFTING FROM A PARADIGM WHERE WE THINK ABOUT ONE ORG NAME WE FIND IN NATURE THAT IS EVOLVING TO MAKE ONE PARTICULAR TYPE OF MEDICINAL COMPOUND TO INSTEAD, LEVERAGING THE VAST BIODIVERSITY THAT IS AVAILABLE TO US IN NATURE. AND SPECIFICALLY, BASED UPON ADVANCES IN DNA SEQ WHERE MORE AND MORE PEOPLE ARE GOING IN AND SEQUENCING DIFFERENT MEDICINAL PLANTS AND ALSO MANY OTHER DIFFERENT ORGANISMS AND THEN DES POSITING THAT SEQUENCE INFORMATION TO US AS RESEARCHERS WITH DIFFERENT COMPUTATIONAL TOOLS AND BEGIN TO HARVEST THOSE GENOMES AND IDENTIFY ACTIVITIES THAT WE THINK ARE GOING TO BE USEFUL OR IMPORTANT IN TERMS OF BUILDING UP MOLECULAR SCAFFOLDS. WE CAN THEN TAKE THE TOOLS OF SYNTHETIC BIOLOGY IN A SYNTHESIS TO TAKE THOSE IDENTIFIED ACTIVITIES AND BASICALLY RECODE THEM SO THAT WE CAN MOVE THEM INTO OUR MICROBIAL HOST AND SPECIFICALLY RECODE THEM SO THAT THEY WILL EXHIBIT ACTIVITIES WITHIN THIS VERY DIFFERENT HOST. AND THEN FINALLY, AS WE HAVE THESE YEAST STRAINS, WE HAVE A PRODUCTION PLATFORM THAT WE CAN GROW VERY QUICKLY AND SHIFT THEM FROM A GROWTH CYCLE, ANNUAL GROWTH CYCLE, TO ONE IN WHICH WE CAN GROW OUR MICROBES OVER A PERIOD OF DAYS AND GROW THEM AT DIFFERENT SCALES AND BASICALLY THEN BE ABLE TO SOURCE MANY, MANY DIFFERENT TYPES OF MEDICINAL COMPOUNDS AND GO BEYOND IMPORTANTLY, WHAT IS AVAILABLE TO US IN NATURE BECAUSE WE ARE ABLE TO FIND THE DIFFERENT BIO50s DIFFERENT SOURCES WITHIN A SINGLE HOST AND PUT THEM TOGETHER IN WAYS YOU WOULDN'T FIND IN NATURE. AND ONE OF THE THINGS THAT WE FELT WAS VERY FIT BEING THIS TYPE OF APPROACH WAS WE COULD BEGIN TO LEVERAGE WHAT WE WOULD CALL DIVERSITY ORIENTED BIOSYNTHESIS AND THE WAY WE CAN DO THIS IS BASICALLY LEVERAGING THE ARCHITECTURES OF THESE PATHWAYS AS THEY ARE FOUND IN NATURE AND SO SPECIFICALLY IF WE THINK ABOUT THE ARCHITECTURES OF THESE PATHWAYS, WE CAN BREAK THEM INTO THE INITIAL BUILDING BLOCKS THAT THEY START WITH WHICH OUR HOST WILL PROVIDE LIKE AMINO ACIDS, THESE BASICALLY THEN GET BUILT-UP THROUGH A COMMON SET OF ENZYMES AND PROTEINS INTO ADVANCED SCAFFOLD LIKE THE ONE I SHOWED HERE, AND THEN THERE IS A SERIES OF DOWNSTREAM ENZYMES WHICH WILL TAILOR THAT SCAFFOLD IN THIS EXAMPLE TO THE ALKALOID. BUT, YOU CAN TAKE THAT SAME ADVANCED INTERMEDIA AND THE BY MODIFYING THE DOWNSTREAM ENZYMES THAT TAILOR THAT ADVANCED INTERMEDIAIAT, YOU CAN ACCESS A BROAD DIVERSITY OF ALKALOID SCAFFOLDS YOU SEE HERE. SO SPECIFICALLY, YOU GO FROM THE ALKALOIDS TO OTHER COMPOUNDS LIKE THESE AND EVEN THE POINT YOU CAN GET TO THIS DIVERSITY BY LEVERAGING A LOT OF THE UPFRONT RESEARCH HAVE YOU DONE TO BUILD OUT PLATFORMS TO MAKE ADVANCED SCAFFOLDS AND THEN DO TAILORING AT THE END. SO THROUGH THE COURSE OF THE PIONEER AWARD WE HAVE BEEN ABLE TO SHOW THAT WE CAN IN FACT RECONSTITUTE THE PATHWAYS TOWARDS COMPOUNDS THAT YOU SEE AND WE HAVE BEEN ABLE TO BUILD OUT THE ADVANCED INTERMEDIATE AND ABOUT 70% OR THREE-QUARTERS OF THE COMPOUNDS THAT YOU SEE AROUND THE YEAST ALREADY BY LEVERAGING THAT TYPE OF TECHNOLOGY OR APPROACH. AND SO I'M GOING ZOOM IN FOR THE PURPOSE OF THIS TALK ON ONE PARTICULAR STORY PRIMARILY AND TALKING ABOUT THAT, I'M GOING TO HIGHLIGHT FOR YOU CHALLENGES THAT CAME UP AS WE WERE DOING THIS WORK AND BASICALLY THE TOOLS THAT WE HAD TO DEVELOP TO ENABLE THIS TYPE OF ENGINEERING AND PATHWAY RECONSTRUCTION. SO I'LL SPECIFICALLY TALK ABOUT A STORY IN YEAST. YOU HAVE A YEAST STRAIN TO GROW ON SUGAR AND TAKE A MOLECULE LIKE GLUCOSE AND BUILD IT BACK AGAIN TO A SCAFFOLD YOU SEE THERE. THIS YEAST STRAIN HAS TO BE ENGINEERED IN 23 DIFFERENT GENES AND THEY WERE TAKEN FROM FOUR DIFFERENT PLANTS AS WELL AS A MAMMAL AND RAT AND A BACTERIUM. AND YOU'LL SEE ONE OF THE REASONS THAT WE DID THIS IS SO ULTIMATELY THIS YEAST IS DOING SOMETHING DIFFERENT THAN WHAT OCCURS IN THE OPIUM POPPY AND IT'S ABLE TO GO BEYOND THE COMPOUNDS WE GET FROM THE OPIUM POPPY AND PRODUCE THE HIGHER VALUE MEDICINAL COMPOUNDS DIRECTLY. AND SO WHAT WERE THE BARRIERS THAT CAME UP WHEN WE DID THIS RESEARCH? WHY DID IT TAKE US SO LONG? A LOT OF BARRIERS WHEN WE LOOK ACROSS MOLECULES, THEY WERE CONSERVED OF THE AS WE DEVELOPED TOOLS TO ADDRESS THESE BARRIERS, WE WOULD BE ABLE TO APPLY THEM GENERALLY AND EXPAND ON COMPOUNDS. WHEN YOU HAVE PATHWAYS THIS LONG, OFTENTIMES THERE ARE GAPS IN THE KNOWN BIOSYNTHETIC PATHWAYS. SO WAYS TO ACCELERATE DISCOVERY BECOMES VERY IMPORTANT TO DO THIS WORK. THE SECOND IS THAT YOU'RE TAKING ENZYMES THAT HAVE EVOLVED TO OPERATE WITHIN VERY SPECIALIZED ENVIRONMENTS SUCH AS THE MEDICINAL PLANTS AND ASKING THEM THEN TO WORK IN A VERY DIFFERENT ENVIRONMENT THE YEAST HOST, AND A LOT OF DIFFERENCES JUST BETWEEN YEAST CELLS AND PLANT CELLS SUCH THAT WHEN YOU MOVE THOSE ENZYMES OVER INTO THE YEAST CELLS, A LOT OF TIMES THEY WILL BE NON FUNCTIONAL EVEN THOUGH THE PROTEIN IS BEING MADE. AND SOMETHING THAT GOES ALONG WITH THAT IS THAT PLANTS HAVE A PREVALENCE OF ENZYMES LIKE CYTOCHROME P450s WHICH NEED TO BE PROCESSED AND LOCALIZED VERY SPECIFICALLY WITH ENDOMEN BRAINS IN THE PLANTS AND GETTING THOSE PROCESSES TO BE RECAPITULATED WITHIN YEAST AND THOSE ENZYMES FOLDED AND CONTINUES TO REPRESENT A BIG CHALLENGE IN THE FIELD. SO THESE ARE CHALLENGE THAT IS CAME UP IN THE COURSE OF OUR WORK AND SO, I'M GOING TO AGAIN START AND HIGHLIGHT NOT EVERY PATHWAY BUT WHERE THESE CHALLENGES COME UP AND THE APPROACHES WE USED. AND IF WE START AT THE FIRST PART OF THE PATHWAY WITH SUGAR AND THIS IS A PATHWAY DIAGRAM WHERE THE AIROISE WILL REPRESENT ENZYMES ACTING AS CATALYSTS FOR A PARTICULAR STEP, AND YOU CAN SEE THE COMPOUNDS THERE. AND SO, IN REMEMBERS IT OF BUILDING UP TO THIS ADVANCED INTERMEDIATE RETIC LIN, WHEN WE STARTED THIS PROECT JUDGE, THE OPIUM POPPY AND THE PLANT BUILT THIS COMPOUND WAS NOT KNOWN. SO WE HAD TO FIGURE OUT HOW TO GILL THE GAP IN THIS PATHWAY. IN THIS PARTICULAR INSTANCE WE DECIDED TO LOOK BROADER BEYOND THE OPIUM POPPY AND PULL OUT ENZYME ACTIVITIES FROM OTHER ORGANISMS. AND IN PARTICULAR, WE COULDN'T DO THAT BECAUSE OF THE BUILDING BLOCKS LIKE DOPAMINE, WE KNOW MAMMALS ARE VERY GOOD AT MAKING DOPAMINE SO WE COULD PULL OUT MAMMALIAN GENES AND IN THIS CASE, WE TOOK GENES FROM THE RAT TO MAKE L DOPA IN OUR YEAST AND THEN WE USED BACTERIAL DECARBOXYLATE TO CONVERT TO DOPAMINE AND WE DID THAT EFFICIENTLY. SO THAT'S WHERE THE MAMMALIAN AND BACTERIAL GENES COME INTO PLACE IN TERMS OF FILLING THAT GAP IN OUR KNOWLEDGE OF THE OPIUM POPPY PATHWAY. AND THEN WE COULD USE DIFFERENT MEDICINAL PLANT GENES TO BUILD OUT. WHEN WE DID THIS, WE HAD A YEAST STRAIN THAT PRODUCED RETIC LIN ALTHOUGH VERY SMALL QUANTITIES AND SO THEN WHAT WE DID WAS A NUMBER OF DIFFERING ENGINEERING MODIFICATIONS TO INCREASE THE AMOUNT OF RETIC LIN BEING BUILT AND ESSENTIALLY WHAT WE DID WAS MADE DIFFERENT MODIFICATIONS TO THE UPSTREAM PART OF THE ENDOGENOUS YEAST METABOLISM PUTTING IN MUTATIONS INTO ENZYMES THAT WERE NATURALLY WITHIN YEAST REMOVE FEEDBACK INHIBITION, THOSE TYPES OF REGULATORY STRATEGIES, THAT ESSENTIALLY ALLOWED US TO INCREASE THE AMOUNT OF RETIC LIN BEING PRODUCED BY 5-FOLD. AND WITH THAT AMOUNT OF RETIC LIN, WE COULD GO ON AND BUILD OUT THE REST OF THE PATHWAY THAT YOU SEE HERE. AND SO AGAIN, I'M GOING TO FOCUS ON A COUPLE OTHER VIGNETTES IN TERMS OF CHALLENGE THAT IS CAME UP. AND ONE OF THE CHALLENGE THAT IS CAME UP THAT WE HAD TO BUILD WAS BASICALLY ACCOUNTING FOR THE DIFFERENCES IN HOW PLANTS HAVE EVOLVED TO PERFORM THEIR METABOLISM RELATIVE TO WHAT YOU MIGHT FIND IN A MICROBE AND IN PARTICULAR, ONE OF THE THINGS WE WERE INTRIGUED ABOUT WAS PLANTS ARE MULTICELLULAR ORGANISMS AND HAVE SPECIFIC TELAND TISSUE TYPES AND MANY CASES WITH SPECIALIZED METABOLISM THEY WILL SEPARATE PATHWAY LIKE WHAT I SHOWED YOU AND SEPARATE ENZYMES AND DIFFERENT PARTS OF THE PATHWAY ACROSS DIFFERENT CELL TYPES AND SHUTTLE INTERMEDIATES BETWEEN THE CELL TYPES AND THERE ARE DIFFERENT REASONS YOU CAN IMAGINE A PLANT EVOLVED TO DO THAT BUT WONG OF THE REASONS COULD BE TO BASICALLY PROVIDE SPECIALIZED BIOCHEMICAL ENVIRONMENT FOR CERTAIN TYPES OF REACTIONS. NOW IF YOU THINK ABOUT A MICROBIAL EXPRESSION STRATEGY OFTENTIMES JUST EXPRESSING THE ENZYMES WITHIN THE CELL AND SO ALL OF THE ENZYMES CAN SEE EACH OTHER SIMULTANEOUSLY. BUT IF WE THINK AGAIN ABOUT YEAST, WE KNOW THAT YEAST DOES HAVE ENDOMEMBRANE COMPARTMENTS AND WE CAN BEGIN TO BASICALLY THINK ABOUT MIMICKING THE SPACIAL REGULATORY STRATEGIES THAT PLANTS HAVE WITHIN A SINGLE YEAST CELL BY LEVERAGING THE ENDOMEMBRANE COMPARTMENTS. WHAT WE DID WAS DEVELOPED A SET OF TAG THAT IS WE COULD USE TO RECODE THE TERMINIS OF OUR ENZYME THAT WOULD BASICALLY DIRECT THOSE ENZYMES TO DIFFERENT ORGANELLES WITHIN OUR YEAST HOST. AND IN DOING, THIS WE WERE GOING TO USE THIS STRATEGY TO ACHIEVE BIOCHEMICAL SPECIALIZATION AND COM PAST MENTALLIZATION WITHIN OUR PATHWAY. SO I'LL GIVE YOU AN EXAMPLE OF HOW WE USED THIS. THIS IS ZOOMING IN ON SOME REACTION STEPS WITHIN THE PATHWAY AND ONE THING THAT IS COMMONLY FOUND WHEN YOU PUT THESE PATHWAYS INTO A MICROBE AND OUT OF THE PLANT HOST, IS THAT YOU BEGIN TO GET PROMISCUOUS ACTIVITY AND IN THIS CASE, THESE TWO ENZYMES COR AND CODM COULD ACT ON TWO INTERMEDIATES OF THE PATHWAY. SO WE ESSENTIALLY USED THOSE RECODING TAGS TO BEGIN TO BASICALLY DELOCALIZE THE ENZYMES WITHIN THE CELL SO WE COULD DIRECT WHAT INTERMEDIATES WITHIN THE PATHWAY THOSE ENZYMES WE ARE SEEING AND WHEN WE DID THAT APPROACH, WHAT YOU COULD SEE IS THAT IS THIS THE BASE CASE WHERE ALL THE ENZYMES ARE EXPRESSED IN THE CYTOSOL AND THIS IS THE YEAHS WE LOCALIZE THAT ENZYME TO DIFFERENT COMPARTMENTS WITHIN THE YEAST THAT WE ARE ABLE TO CHANGE THE SPECIFICITY BETWEEN THE AMOUNT OF MORPHINE AND NEOMORPHINE PRODUCED AND IN SOME DESIGN, ABLE TO GET IT UP TO 90-95% SPECIFICITY TOWARDS THE DESIRED PRODUCT. NOW THE OTHER THING WE HAD TO DO WAS DO SOME ENZYME DISCOVERY AND SO ZOOMING IN PART OF THE PATHWAY, THERE WAS A GATEKEEPER ENZYME WHICH BASICALLY IS THE ONE THAT SPECIFICALLY DIRECTS FLOW TOWARDS THE CALL LLOYDS WHICH BASICALLY DOES A REACTION FROM RETICK LIN TO R RETICK LIN. WE COULDN'T GO INTO SORT OF ORGANISMS LIKE WE DID WITH THE EARLY PART OF THE PATHWAY BECAUSE THIS IS GOING TO BE VERY SPECIFIC FOR PLANTS THAT MAKE MORPHINE. SO WHAT WE DID IS, WE WEPT INTO GENOME DATABASES FOR THESE MEDICINAL HAPPENS AND DEVELOPED A CHEMICAL HYPOTHESIS FOR WHAT TYPE OF ENZYME WOULD DO THE SECOND REACTION, REDUCTASE ENZYME AND DID A HOMOLOGY SEARCH TO SEARCH FOR ENZYMES THAT ARE PRESENT IN POPPIES THAT MADE MORPHINE BUT NOT PRESENT WITHIN RELATED POPPIES THAT DID NOT. WE HAD A FEW HITS THAT CAME OUT OF THIS. ONE OF THE THINGS THAT WAS INTRIGUING IS THAT THEY REPRESENTED THIS VERY UNIQUE NOVEL ENZYME WHICH WAS THE FUSION BETWEEN A PLANT CYTOCHROME P450 AND THE REDUCTASE ENZYME WE WERE LOOKING FOR. AND THIS STOOD OUT TO US BECAUSE THERE HAD NOT BEEN IDENTIFIED A FUSION PROTEIN OF A PLANT-TYPE CHROME P450 AND IN FACT IA CYTOCHROME P450 COULD PERFORM THAT FIRST STEP. THE FIRST OXIDATION OR FIRST STEP OF THAT REACTION TO DEHYDRORETIC LIN. WE SYNTHESIZED A BUNCH OF THOUGHTS CAME OUT OF OUR SEARCH AND PUT THEM INTO OUR YEAST HOST MAKING S RETIC CLIN WE COULD SHOW WHEN WE PUT THAT FUSION ENZYME INTO OUR YEAST HOST IT WAS ABLE TO THEN CONVERT S TO R AND THIS WAS A NOVEL DISCOVERY BOTH IN TERMS OF THE CLASS OF ENZYME AND ALSO BECAUSE THIS WAS A REMAINING STEP OF THE OPIOID SYNTHETIC PATHWAY THAT REMAINED UNDISCOVERED AFTER DECADES OF RESEARCH THROUGH TRADITIONAL PLANT BIOCHEMICAL TECHNIQUES. AND SO THIS HIGHLIGHTS HOW THIS SYNTHETIC APPROACH CAN HAVE A LOT OF IMPLICATION FOR DISCOVERY OF NEW ENZYMES. THE FINAL THING IS THE CHALLENGES THAT COME UP WITH YOU HAVE THESE ENZYMES AND YOU GET THEM TO BE MADE WITHIN THE PLANTS BUT AGAIN BECAUSE THEY ARE BEING MADE IN A VERY DIFFERENT ENVIRONMENT, YOU OFTENTIMES HAVE ISSUES OF ACTIVITY. THIS IS THE CASE THAT CAME UP TO THE CYTOCHROME P450 AFTER THE REACTION WHERE WE KNEW THE ENZYME AND GETTING ACTIVITY ABOUT 1% ACTIVITY. WHEN WE LOOKED ON A PROTEIN IN EXPRESSION IN YEAST, IT LOOKED LIKE IT WAS BEING MADE BUT WHEN WE COMPARED IT TO EXPRESSION OF PLANTS, WHAT WE SAW WAS DISBANDING PATTERN. THAT IS INDICATIVE OF MISGLYCOSYLATION. SO WE COULD IDENTIFY THE PROTEIN WAS BEING NADE BUT MISGLYCOSYLATED AND INCORRECTLY PULT INTO THE MEMBRANE. WE DEVELOPED A NOVEL FUSION STRATEGY WHERE WE BASICALLY RECODED THE END TERMINAL OF THAT PROTEIN SNOW WOULD BE CORRECTLY PROCESSED PROCESSED AND MADE A DUNK OF VARIANTS TO FIGURE OUT WHERE THE BEST PLACE WAS TO MAKE THAT TRUNCATION AND FUSION WE WERE ABLE TO INCREASE ACTIVITY OF PROTEIN BY 10 FOLD AND INCREASE THE PROCESSING. SO, AT THE END OF THIS WHEN WE PUT EVERYTHING TOGETHER, WE WERE ABLE TO DEMONSTRATE WE HAD YEAST THAT COULD GROW ON SUGAR AND OVER THE MATTER OF THREE DAYS MAKE COMPOUNDS LIKE HYDROCODONE. ONE OF THE THINGS WE WERE VERY INTRIGUED IS THE ABILITY TO DO VERY RAPID DIVERSITY ORIENTED BIOSIN THIS. ONCE WE DID THIS WORK AND PROJECT, I SAID IT TOOK MANY, MANY YEARS TO ADDRESS ITS PROBLEM, WE WERE ABLE TO LEVERAGE THE TOOLS THAT I JUST DESCRIBED AND THE PLATFORMS AND INSTEAD MAKE A VERY DIFFERENT BUT RELATED ALKALOID COMPOUND AND THIS WORK TOOK ONE POSTDOC LESS THAN A YEAR. SO YOU CAN SEE THE EFFICIENCY THAT IS GAINED NOW THAT WE HAVE THESE TOOLS IN PLACE AND MANY OF THESE PLATFORMS IN PLACE. AND SO, I'LL JUST END BY SAYING ONE OF THE THINGS WE ARE VERY EXCITED ABOUT WITH THIS WORK AND WITH THESE PROTEIN CAPABILITIES IS REALLY BEING ABLE TO TAKE IT AND APPLY IT VERY BROADLY TO THE SYNTHESIS OF COMPOUNDS THAT ARE INSPIRED BY THOSE THE PLANTS MAKE BUT GO BEYOND WHAT WE FIND IN NATURE TO BUILD BETTER COMPOUNDS AND BETTER MEDICINES. AND WITH THAT, I WILL THANK MY LABORATORY WHO IS RESPONSIBLE FOR THE WORK. I'LL ALSO THANK THE NIH, COMMON FUND, AND ALSO NIH WHO FUNDED MY PIONEER AWARD AND REALLY WE HAD A LOT OF TROUBLE GETTING THE WORK FUNDED FRIAR TO THAT. IT WAS DEEMED AS INCREDIBLY RISKY. I HAD A LOT OF REVIEWS THAT SAID WHAT WE WERE THIGHING TO DO WAS IMPOSSIBLE. SO HAVING THIS TYPE OF FUNDING AND SUPPORT FOR THIS WORK REALLY ALLOWED US TO PUSH ON IT AND DO THINGS WHICH WE WEREN'T ABLE TO DO BEFORE THAT AND NOW THAT WE HAVE BEEN ABLE TO DEMONSTRATE THIS IS POSSIBLE, THERE HAS BEEN A LOT OF OTHER PROJECTS AND FUNDING FUNDING FUNDING THAT HAVE COME AFTER THAT. WE HAVE WORK NOW WE ARE VERY EXCITED ABOUT WHERE WE ARE TAKING THE APPROACHES I DESCRIBED AND DIRECTING THEM TOWARDS DISCOVERY OF NEW PATHWAYS AND NEW NATURAL PRODUCTS WHICH ARE BEING FUNDED THROUGH NIGMS. SO THANK YOU. [ APPLAUSE ] HAPPY TO TAKE QUESTIONS IF THERE ARE ANY. >> IS THIS A REALLY COOL -- I HAVE A NAIVE QUESTION, GENERAL KIND. SO YOU WERE MENTIONING THE SUPPLY CHAIN PROBLEMS WITH POPPY. BUT IF MAKING MORPHINE OR OXYCODONE BECOMES AS EASY AS BREWING BEER THEN PRESUMABLY ALL KINDS OF INTEREST AND CONSEQUENCES. >> GREAT QUESTION. SORT OF ON THE IMPLICATIONSES OF THIS TYPE OF WORK. AND SO, THERE WAS A LOT OF MEDIA THAT DISCUSSED THIS AND WHEN THIS WORK WAS BEING PUBLISHED OVER THE PAST YEAR FROM NYE LAB AND OTHERS AS WELL. ONE OF THE THINGS I WOULD SAY IS THAT WE DID A FOLLOW-UP STUDY TO THE WORK DESCRIBED IN SCIENCE WHERE WE TOOK OUR YEAST AND WE GREW THEM AS YOU WOULD FERMENT BEER. AND SO, WHAT WE SHOWED IS THAT -- AND SO LET ME TAKE A STEP BACK AND SAY THE WAY YOU GROW INDUSTRIAL STRAINS IN THE LABORATORY AND BOTH AT A COMPANY, ARE VERY, VERY DIFFERENT THAN HOW WE BREW BEER, MEANING THERE IS SO MUCH CONTROL OVER HOW YOU FEED THE SUGAR, WHAT YOU FEED THEM, HOW YOU CONTROL THE PH, HOW YOU CONTROL AERATION, ET CETERA. IT'S VERY TIGHTLY CONTROLLED AND ALL OF IT HAS -- AND THAT HAS VERY SIGNIFICANT IMPLICATIONS FOR HOW MUCH PRODUCT IS BEING PRODUCED. AND WHAT WE SHOWED IN THE FOLLOW-UP PAPER WAS THAT WHEN WE TOOK THE SAME STRAINS AND BASICALLY BREWED THEM AS BEER, WE SAW NO PRODUCTION OF OPIOIDS. A LITTLE BIT OF PRODUCTION OF RETIC LIN BUT NOTHING BEYOND THAT. ESPECIALLY AS WE CONTINUE TO IN PROVE STRAINS THAT GO INTO COMMERCIAL PRODUCTION, IF YOU PUT THEM IN A BEER BREW KILT THAT YOU WOULD NOT SEE ACCUMULATION BUT IT WILL BE MUCH, MUCH SLOWER THAN WHAT YOU WOULD GET IN INDUSTRIAL OPTIMIZE PROCESS AND THEN YOU WOULD REALLY HAVE TO ASK YOURSELF IF THAT IS A FEASIBLE TIME TO GET COMPOUNDS GIVEN THERE ARE OTHER WAYS TO DO THAT IF THAT'S WHAT YOU WERE INTERESTED IN. YOU JUST TAKE A HUGE HIT IN INEFFICIENCY. >> SO YOU WERE TALKING ABOUT GROWING, WHETHER THERE IS A DIFFERENCE -- TO CONTROL THE CELL CYCLE. DO YOU SYNCHRONIZE THE CELLS? AND IN TERMS OF MEMBRANE LOCALIZATION OR COMPARTMENTALIZATION MIGHT THAT BE A WAY TO IMPROVE YOUR EFFICIENCY? >> YES. THAT IS A GOOD QUESTION. SO, FIRST PART OF YOUR QUESTION, DO WE CONTROL CELL CYCLE. EFFECTIVELY WHEN WE ARE GROWING THESE FOR PRODUCTION, NO. BECAUSE WHEN YOU GROW THESE IN INDUSTRIAL FERMENTER, YOU BASICALLY MOST OF THE TIME WILL BE FED BATCH WHERE YOU START WITH THE CELLS LOW INOCULUM AND YOU GROW AT BIOMASS AND THEN KEEP PRODUCTION GOING SO YOU'RE CONTINUING TO PRODUCE AS BIOMASS IS GROWN AND THEN AFTER THE 3-4 DAYS YOU'RE GOING TO BASICALLY HIT THE END OF THE PRODUCTION AND DRAIN OUT THE FERMENTER. SO FROM THE PURPOSE OF INDUSTRIAL PROCESS, YOU DON'T DO ANYTHING TO CONTROL CELL CYCLE OR SYNCHRONIZE THE CELLS. THAT WOULD BE MUCH TOO COMPLICATED AND PROBABLY EXPENSIVE. I AGREE WITH YOU THOUGH FROM THE PERSPECTIVE OF THINKING ABOUT LOCALIZING THE ENZYMES TO DIFFERENT MEMBRANES AND HOW THAT RELATES TO THE GROWTH STAGE OF THE CELL. THIS DATA COULD BE VERY INTERESTING IN TERMS OF LOOKING AT HOW EFFECTIVE THOSE PROCESSES ARE BECAUSE YOU'RE RIGHT. YOU WOULD SEE A LOT MORE EFFICIENCY IN THAT IF YOU'RE SYNCHRONIZING THE CELL CYCLE AND DOING IT AT A SPECIFIC STAGE. >> MAYBE CONTROL CELL CYCLE GENETICALLY -- [ OFF MIC ] >> YOU DEFINITELY COULD AND IT WOULD BE THE SORT OF BALANCE BETWEEN -- YOU VALUES A BALANCE INDUSTRIAL IN TERMS OF LIKE GETTING BIOMASS AS QUICKLY AS POSSIBLE BECAUSE THOSE BECOME YOUR CELL CAT LEFT AND HOW THAT WOULD EFFECT THE OVERALL PROCESS. >> CHRISTINA, THE PROCESS OF MAKING A BRIDGED ALKALOID FASCINATES ME. I'M GOING TO GET THE CHEMISTRY WRONG -- PSYCHE LOW COMPOUNDS. I DON'T THINK ANIMALS CAN MAKE THOSE, ONLY PLANTS DO. SO, IS THERE -- DID YOU NEED TO PUT THESE IN A SPECIAL ORGANELLE? WHAT IS THE COMB INDUSTRY AND ENZYMOLOGY? >> I MISSED THE FIRST PART OF YOUR QUESTION. >> BRIDGED ALKALOIDS. IS THIS ESPECIALLY DIFFICULT TASK? >> TO PUT IT INTO ORGANELLES? >> TO MAKE THEM. >> TO MAKE THEM? I THINK DIFFICULT IS A LITTLE BIT RELEVANT. >> IS THERE SELECTEDDIVE ADVANTAGE TO MAKING A BRIDGED ALKALOID. WE DON'T MAKE THEM AS FAR AS I KNOW. >> I THINK PLANTS DO DO CERTAIN CHEMISTRIES REALLY AND WILL THEY ALSO -- IF YOU LOOK AT THE ENZYMES PRESENT, THERE IS JUST AN ABUNDANCE OF PLANTS WHICH DO VERY DIFFERENT CHEMISTRIES BEYOND OXIDATION ESPECIALLY WITHIN PLANTS. AND I THINK THAT IS WHERE YOU GET A LOT OF THE RICHNESS AND THE SCAFFOLDS BUILT-UP AS WELL AS OTHER ENZYMES AS WELL. SO, I THINK SO IS THERE ANY REASON WHY WE DON'T SEE IT IN OTHER ORGANISMS YOU DON'T SEE THE -- AS PREVALENT AS THE TYPE OF DIVERSITY OF ENZYMES YOU WOULD SEE IN PLANTS. THAT IS OFTEN OVER REPRESENTED WITHIN THOSE PATHWAYS. AGAIN, SOME PEOPLE NEED TO HAVE MORE DEDISPENSE THEY JUST EVOLVED TO DO THOSE TYPES OF CHEMISTRIES. >> TAKE ME BACK TO SQUARE ONE. CYTOCHROME B450 DOING THIS? >> IT DOES. -- IN THOSE PATHWAYS I HAVE TO LOOK AT IT. THERE IS -- LIKE IN THE MORPHINE PATHWAYS SOME OF T CYCLIZATION AND SOME OF THE CARBON CARBON COUPLINGS DOES HAPPEN THERE. 13469 OTHER ONES I HAVE TO LOOK TAT AND SEE. I HAVEN'T STUDIED THOSE AS CLOSELY BUT IT'S POSSIBLE. [ APPLAUSE ] >> OUR NEXT SPEAKER IS LEOR WEINBERGER WHO WILL TALK ABOUT A HARDWIRED HIV LATENCY PROGRAM. >> SO, I'M GOING TO MOTIVATE HIV LATE OPENSY VIRUS MINDING YOU ABOUT THE FRESHMAN CHEMISTRY. YOU PROBABLY ALL REMEMBER OF THE URANNUOUS EQUATION. AND IT DESCRIBES TRANSITIONS BETWEEN TWO STATES IN A CENTRAL THEORY MODEL. THE NUMERATOR DESCRIBES THE HEIGHT OF THE ACTIVATION ENERGY BARRIER, CATALYST TO A LOWER HEIGHT OF THE ACTIVATION BARRIER AND FOR CENTURIES CHEMISTS USED BACK UPS AND BURNERS TO INCREASE THE THERMAL FLUCTUATIONS, DENOMINATOR. BIOLOGY HAS BEEN DOING SOMETHING SIMILAR FOR A LOT LONGER. OKAY, SO I'LL TELL YOU ABOUT A HARDWIRED OR FLUCTUATION DRIVEN LATENCY PROGRAM IN HIV. THE VIRUS INFECTS CD4 CELLS AND CAN DECIDE TWO ALTERNATE STATES, ACTIVE REPLICATION AND DORMANT OR LATENT STATE AND I'LL SHOW YOU DRIVEN BY FLUCTUATIONS MUCH THE SAME WAY THAT FLUCTUATIONS DRIVE BETWEEN TWO DIFFERENT STATES IN CHEMISTRY. SO THE FIRST PART OF THE TALK, THE OUTLINE, I'LL TELL BUT THIS NOISE-DRIVEN CIRCUIT IN THE VIRUS AND I WANT TO WARN YOU I TEND TO GIVE LONG INTRODUCTIONS SO IF EIGHT MINUTES IN WE STILL DON'T HAVE DATA, DON'T BE TOO ALARMED, I'LL GET THERE. AND WE CAN TARGET THIS PHENOMENON, FLUCTUATIONS, AS A THERAPEUTIC. USE THEM, HARNESS THEM, EXPLOIT THEM FOR THERAPEUTIC POTENTIAL. SO, I WANT TO THANK THE PEOPLE WHO DID THE WORK UPFRONT BECAUSE SOMETIMES I DON'T FINISH. I'M GOING PRIMARILY FOCUS ON THE WORK OF TWO PEOPLE, ONE IS ROY DARR WHO IS A POSTDOC IN THE LAB RECENTLY STARTED HIS OWN LAB AND ANOTHER EXTREMELY TALENTED POSTDOC GOING ON THE JOB MARKET NEXT YEAR AND A LOT OF THIS WORK WAS DONE UNDER THE NEW INNOVATOR AWARD AND SOME OF IT UNDER THE PIONEER AWARD. SO I WILL TELL YOU ABOUT FLUCTUATIONS. WHERE ARE THE FLUCTUATIONS COMING FROM? THEY ARE OBVIOUSLY NOT THERMAL LIKE IN CHEMISTRY. BIOLOGY OPERATES IN A VERY NARROW REGIME. YOU PUT A CELL IN A BURNER AND YOU KNOW WHAT HAPPENS. INSTEAD THE FLUCTUATIONS ARE COMING FROM RANDOM OR DIFFUSION-DRIVEN PROCESSES. SO GO BACK TO CENTRAL DOGMA. ALL THESE ARROWS ARE ENZYME MEDIATED EVENTS. THE ENZYMES HAVE TO COLLIDE WITH SUBSTRATES INSIDE THE CELL. SO THE RIBOSOME HAS TO FIND THE RNA SUBSTRATE ET CETERA. AND BECAUSE THESE ARE DIFFUSION LIMITED PROCESSES AND THERE ARE TYPICALLY FEW BIOREACTANTS INSIDE THE CELL, IF YOU LOOK AT INDIVIDUAL CELLS OVER TIME SHOWN HERE, YOU DON'T SEE THIS GRAY SMOOTH LINE YOU SEE IN BIOCHEMISTRY TEXTBOOKS. INSTEAD YOU SEE JAGGED INCREMENTS OF INTEGER NUMBERS OF MOLECULES AND THIS IS MOST BEAUTIFULLY SHOWN BY THIS LAB. SO, YOU CAN SEE REGIONS WHERE LITTLE IS OCCURRING BECAUSE THE DIFFUSION LIMITED EVENTS ARE DIFFICULT TO FIND THE ENZYMES AND THERE IS REGIONS WHERE YOU HAVE A LOT OCCURRING IN A SMALL AMOUNT OF TIME. IF YOU TAKE A GEFELTICALLY IDENTICAL CELL WHICH IS EVEN DIVIDED FROM THIS SAME CELL, YOU CAN GET A DIFFERENT RANDOM TRAJECTORY BASED OFF OF THESE STOCHASTIC OR DIFFUSION LIMITED EVENTS. THE GRAY LINE IS EXTRACTION. NO CELL TO FOLLOW THIS GRAY LINE IT'S IDEALIZATION YOU GET FROM DIVIDING MANY HUNDREDS OF THOUSANDS OF CELLS TOGETHER. YOU KNOW THIS BECAUSE IF YOU TAKE A TIME SLICE FOR INSTANCE OVER HERE, WE DO EVERY DAY SPLIT TOM TREE, YOU GET A HISTOGRAM. THERE ARE MANY CELLS WHICH ARE TRANSITING THROUGH THE IDEALIZED MEAN OVER HERE BUT THERE ARE ALSO MANY CELLS TAKING LONG EXDISCUSSIONS FROM THE MEAN. THE WIDTH OF THE DISTRIBUTION WHAT WE CALL THE NOISE. WE HAVE MANY METRICS FOR IT. STANDARD DEVIATION, I'LL BE USING THE COEFFICIENT VARIATION WHICH IS THE NORMALIZED STANDARD DEVIATION. AND TYPICALLY WE THINK OF THIS AS A NUISANCE, SOMETHING BIOLOGY HAS TO MINIMIZE BECAUSE IT'S A DETRIMENT BUT ABOUT 50 YEARS AGO, THERE WAS A THEORETICAL ONCOLOGIST, DAN COHEN, WHO WAS STUDYING DESERT ANNUAL PLANTS AND MADE AN OBSERVATION THAT A SINGLE PLANT COULD PRODUCE SEED THAT IS HAVE THICK HUSKS AND THIN HUSKS, BASICALLY THIS VARIABILITY. AND TO EXPLAIN THIS, HE MADE THE FORMAL MATHEMATICAL ANALYSIS TO FINANCIAL PRACTICE -- DOING FOR DECADES IN A VOLATILE MARKET. PUTTING SOME ASSETS INTO LONG TERM SAVINGS ACCOUNTS TO GUARD AGAINST MARKET CRASHES AND SOME INTO HIGH-YIELD STOCKS. AND COHEN SAID THE PLANTS WHICH HAVE TO CONTEND THESE ANNUALS HAVE TO CONTEND WITH UNCERTAIN DROUGHT-LIKE CONDITIONS WERE DOING THE SAME THING. SOME OF THEIR SEEDS WENT INTO THICK HUSKS WHICH DIDN'T GERMINATE VERY EASILY BUT DIDN'T DEFECATE EASILY IN A DROUGHT AND SOME WENT INTO THIN HUSKS WHICH JEOP GERMINATED VERY EASILY WITH MOISTURE. AND BY VIRTUE OF THIS VARIABILITY YOU COULD HAVE TWO DIFFERENT PHENOTYPES. THIS IS THE ORIGINS OF DEVELOPMENTAL BED HEDGING THEORY. THE BEAUTIFUL THEORY IS THAT ESSENTIALLY THE ORGANISMS COULD MAXIMIZE THEIR FIT INNOCENCE A FLUCTUATING ENVIRONMENT BY MATCHING THEIR VARIABILITIES TO THE ENVIRONMENTAL VARIABILITIES. IT WAS A BEAUTIFUL THEORY. FOR 40 YEARS IT WAS TREATED THAT WAY. A THEORY. WE HAD NO EVIDENCE OF THIS. THE REASON WE DIDN'T HAVE THAT IS BECAUSE WE WERE MISSING THE LEFT-HAND SIDE OF THE PICTURE. WE COULDN'T SEE THE FLUCTUATIONS. THEN IN THE 1990S WITH THE ADVENT OF FLORESCENT PROTEINS WE STARTED TO SEE IT. NOW WE HAVE LOTS OF EXAMPLES OF THIS TYPE OF STOCHASTIC HEDGING BETWEEN ALTERNATE PHENOTYPES. SHEAR A FEW OF THOSE EXAMPLES. I'LL BE TELLING YOU ABOUT THE FIRST ONE, HIV LATENCY, A FEW YEARS AFTER WE PUBLISHED THIS BEAUTIFUL WORK FROM MICHAEL ELLIS'S LAB SHOWING THAT NOISE OR FLUCTUATIONS IN GENETIC CIRCUIT AND BACTERIA CAN DRIVE TWO DIFFERENT PHENOTYPES. AND THEN A NUMBER OF OTHER LABS, STEM CELLS, AND CANCER CELLS AND MANY OTHER SYSTEMS NOW. AND THESE ARE ALL EXAMPLES OF CELLULAR DECISIONS IN THESE EPIGENETIC LANDSCAPES I'LL SHOW YOU HIV HAS TWO STATES, OFF STATE AND ON STATE AND FLUCTUATIONS IN GENETIC CIRCUIT ALLOW THE VIRUS TO TRANSITION BETWEEN TWO STATES AND ACTUALLY THIS IS INTRINSIC PROGRAM WE CAN THERAPEUTICALLY TARGET AND PUSH THE VIRUS BETWEEN THESE TWO DIFFERENT SPACES. SO, TO MOTIVATE HIV DORMANCY OR LATENCY, HERE IS ONE EXAMPLE, WELL-KNOWN EXAMPLE OF THE MISSISSIPPI BABY. A CHILD WAS TREATED VERY AGGRESSIVELY AFTER BIRTH BUT THEN LOST TO FOLLOW-UP, DECLARED CURED BUT WHEN THE CHILD CAME BACK 18 MONTHS LATER AFTER FOLLOW-UP, WAS FOUNDS TO BE HIV POSITIVE. IN EVERY PATIENT THAT THERAPY HAS BEEN DISCONTINUED IN, THE VIRUS HAS REBOUNDED. SO THIS IS THE CHIEF BARRIER WE HAVE TO CURE PATIENTS. HERE IS THE UNDERLYING PHENOTYPE. HIV INFECTS CD4 POSITIVE T-CELLS, THE VIRUS IS AN RNA VIRUS WHICH REVERSE THE RNA INTO DNA. PSEUDORANDOMLY INTEGRATES THAT DNA INTO THE HOST GENOME LIKE GENES BUT CAN GO ANYWHERE INSIDE THE GENOME AND THEN RAPIDLY DESTROYS THE CELL IN 40 YOURS PRODUCING THOUSANDS OF VIRAL PROGENIES. SOMETIMES WHEN THE VIRUS INFECTS A CELL, THE PROGRAM EITHER SILENCES OR DOESN'T ACTIVATE AND THIS FORMS A LATENT RESERVOIR WHERE THE CELLS ARE INFECTED BUT NOT ACTIVE AND THEY CAN REACTIVATE AFTER AN AMOUNT OF TIME AND REPEAT THE INFECTION. ALL OF OUR DRUG COCKTAILS TARGET ACTIVE REPLICATION STATE. WHEN PATIENTS ARE REMOVED FROM THERAPY, THE VIRUS CAN REBOUND. NO DOCTORS ARE TARGET THE LATENT CELLS. - AUDIO DANGEUS. THERE IS A THEORY FOR OUR PROPOSAL FOR CURING A PATIENT, WHICH IS TO GET RID OF THE LATENT RESERVOIR BY REACTIVATING IT OR SHOCKING ALL THE LATENT CELLS TO PUSH THEM INTO THE ACTIVE RESERVOIR AND THEN WE CAN TARGET THESE WITH OUR STANDARD OR CONVENTIONAL ANTI-RETROVIRAL DRUG COCKTAILS. THAT'S BEEN EXCEPTIONALLY CHALLENGING. PART OF THE VANE BECAUSE WE HAD A POOR HANDLE ON HOW THIS ORIGINAL DECISION WAS MADE. SO THERE ARE THREE HYPOTHESES FOR HOW THIS DECISION WAS MADE. THE FIRST WAS THERE WAS SOMETHING ABOUT THE VIRAL INTEGRATION SALE. WHERE THE VIRUS INTERESTS GREATS IN THE GENOME DETERMINES WHETHER THE VIRUS TURNS ON AND REPLICATES OR TURNS OFF. THAT IS STUDIED FOR 15 YEARS. THERE IS NO EVIDENCE OF A CORRELATION BETWEEN VIRAL INTEGRATION SITES AND ACTIVE REPLICATION OF LATENCY. THE VIRUS INTEGRATES THROUGHOUT GENES AND 70% OF THE TIME AND 30% OF THE TIME IT CAN INTEGRATE ANYWHERE. IN ANY OF THOSE TWO CASES NO CORRELATION BETWEEN ACTIVE REPLICATION OR LATENCY. THE SECOND HYPOTHESIS FOR MANY INJURIES SOMETHING ABOUT THE CELL. AND THIS I'LL DISCUSS IN JUST TWO SLIDES. BUT, THESE CD4 T-CELLS NEED TO BE ACTIVATED IN ORDER TO BE INFECTED AND WE KNOW IT'S PART OF THEIR PHYSIOLOGY THEY SILENCE OR RETREAT TO A RESTING MEMORY STATE. THE REASON YOU DON'T GET MEASLES TWICE IS BECAUSE YOU HAVE RESTING MEMORY AND AS THEY ARE SILENCING GOING TO A RESTING STATE, THE VIRUS WENT ALONG FOR THE RIDE AND FOLLOWED THEM INTO THE RESTING STATE. AND THAT IS HOW LATENESSESY WAS ESTABLISHED. THERE ARE A LOT OF PROBLEMS WITH THAT THEORY. MOST WERE TEMPORAL. IT TAKES ABOUT TWO WEEKS TO EFFICIENTLY ESTABLISH RESTING MEMORY AND FOR THE CELLS TO SILENCE. INFECTING A MONKEY. LOTS OF OTHER PROBLEMS AS WELL BUT THAT IS IT IN A NUTSHELL. SO THE THIRD HYPOTHESIS WHICH WAS THE MINORITY HYPOTHESIS FOR MANY YEARS, AND IS THE ONE I'M GOING TO TALK ABOUT WAS THAT IT'S SOMETHING ABOUT THE VIRUS. THE VIRUS ITSELF IN TERMS OF PROGRAM THAT ALLOWS IT TO CHOOSE BETWEEN ACT AND I HAVE LATENT REPLICATION. AND I'M GOING TO SHOW YOU THE BOOK ENDS. THE FIRST EXPERIMENTS WE DID AND THE MORE CONVINCING FINAL EXPERIMENTS WE DID A FEW YEARS AGO. SO SHEAR WHAT THE PROGRAM WE THOUGHTED LOOKED LIKE. HERE IS THE VIRAL GENOME. ONE PROMOTER IN THE 5 PRIME LONG TERM REPEAT. IT'S A WEAK PROMOTOR. IT STALLS JUST DOWNSTREAM ABOUT 70 NUCLEOTIDES DOWNSTREAM OF INITIATION. BUT, SOMETIMES ONE OF THE PROTEIN PATCH SHOWN HERE, NINE GENES OF HIV ARE TURNING GRAY. IT'S TRANSLATED AND FEEDS BACK INTO THE NUCLEUS AND HYPERACTIVATES THE PROMOTOR AND HANGING OFF THE STALLED RNA POLYMERASE. IT'S A TRANSCRIPTION FACT ORAL BUT A VERY DIFFERENT TYPE. REALLY BEAUTIFUL MOLECULAR BIOLOGY SHOWED IT BINDS RNA AND NOT DNA. SO BINDING THAT STALLED RNA POLYMERASE HANKING OFF THE STALLED POLYMERASE HYPERACTIVATING THE CARBOXY TERMINAL DOMAIN BY PHOSPHORYLATING IT AND THE RNA POLYMERASE BECOMES PROCESSIVE. THIS IS A POSITIVE FEEDBACK LOOP. AND FROM ELECTRICAL ENGINEERING, WE KNOW POSITIVE FEEDBACK LOOPS ARE THE EASIEST WAY TO MAKE A SWITCH BETWEEN ON AND OFF. THE AMPLIFY NOISE AND FLUCTUATION AND THEY CAN PRODUCE A SWITCH BETWEEN ON AND OFF. AND SO THE HYPOTHESIS WE HAD WHICH WAS A STRONG HYPOTHESIS, IS THAT THIS IS ENOUGH. JUST PROMOTOR AND IT'S SOFT GIVE YOU THIS DECISION. SO -- SUFFICIENT. SO WE RIPPED OUT JUST THIS CIRCUITRY FROM THE VIRUS AND TESTED WHETHER ITSELF CAN GIVE A DECISION BETWEEN TWO ALTERNATE TRANSCRIPTIONAL STATES. HERE IS THE CONTROL JUST THE LTR DRIVING GFP AND JUST INDIVIDUAL CELLS AT LOW MULTIPLICITY INFECTION, SORT SINGLE CELLS BY BACK AND GREW THEM IN POPULATIONS. HERE ARE NINE DIFFERENT POPULATIONS AND SITES. THERE IS VERY LITTLE VARIABILITY. GFP EXPRESSION ON THIS AXIS AND HISTOGRAM ON THIS AXIS. VERY LITTLE VARIABILITY. THERE IS NOT MUCH CELL TO CELL DIFFERENCE. YOU DO THE EXACT SAME EXPERIMENT WITH THE POSITIVE FEEDBACK LOOPS AND NOW YOU CAN MAKE A GSP FUSION AND DO THIS TRANSCRIPTIONALLY OR TRANSLATIONALLY. SORT A SINGLE CELL WHICH IS SINGLE INTEGRATION OF THIS VECTOR FROM OVER HERE FROM THIS POSITION AND IT BIFURCATES. PHENOTYPICALLY ON AND OFF STATES. DIFFERENT INTEGRATION SITES ARE SHOWN HERE. SOME OF THEM ARE MORE ON THAN OFF. SOME MORE OFF THAN ON. A TEMPORAL ASPECT TO THIS I'LL SHOW YOU IN A SECOND. BUT THIS CIRCUITRY WAS ENOUGH TO GIVE TWO DIFFERENT PHENOTYPIC STATES. THE FIRST HYPOTHESIS TO EXPLAIN THIS WAS VARIEGATION EVEN THOUGH THESE ARE ALL CLONAL POPULATIONS, A SINGLE INTEGRATION CHOSEN FROM HERE, MAYBE THE CHROMATIN IS EXCELLENCED AND THAT'S WHAT IS GIVING YOU THE DIFFERENCE BETWEEN THESE TWO DIFFERENT STATES. SO ONE WAY TO TEST THAT IS TO SEE IF YOU CAN ACTIVATE THE OFF STATE. SO IF ITS POSITION EFFECT HAS TO BE CHROMATIN SILENCING, IT ACTIVATES AND THAT MEANS THE POLYMERASE IS THERE. MANY OF THESE WERE ALSO INSIDE GENES. SHORT STORY, TO MAKE A LONG STORY SHORT, STAT ABLE TOA -- TAT ABLE TO ACTIVATE THESE. HERE IS ONE OF THE MORE RECENT EXPERIMENTS WE D TAKE A FULL-LENGTH VIRUS, KNOCKOUT TA. AND SUPPLY IT IN TRANS. SO IN THIS VIRUS, THERE IS A CHERRY REPORTER IN THE NET BREEDING FRAME. THE CELLS WHICH ENCODE TAT, IT IS EXPRESSED OFF PROMOTOR AND THE EXPERIMENT IS THAT IF YOU TAKE THIS VIRUS, ISOLATE IT AND THEN INFECT CELLS IN THE PRESENCE AND ABSENCE OF TAT OR DOXYCYCLINE. AND THEN YOU MEASURE EFFICIENCY OF INFECTION. SO IN THE PRESENCE OF TAT THERE IS A LOT OF INFECTED CELLS, ABOUT 30%. IN THE ABSENCE VERY LITTLE. THIS DOESN'T PROVE VERY MUCH BECAUSE IT COULD BE ABORTIVE INFECTIONS SO YOU NEED TAT FOR ACTIVE INFECTION BUT WITHOUT IT THE VIRUS COULD HAVE GONE ABORTIVE. WE COULD HAVE DONE THE EXPERIMENT WRONG WHERE WE DIDN'T ADD THE SAME AMOUNT OF VIRUS TO BOTH WELLS. SO TWO DAYS LATER, YOU ADD DOXYCYCLINE AND ALL THE VIRUS COMES BACK. SO IT WAS LATENT AND TAT ABLE TO, IT'S SOFT REACTIVATE IT. THIS IS NOT CONSISTENT WITH CONVENTIONAL CHROMATIN SILENCING OR POSITION EFFECT AT ALL. SO WE HAD A MODEL IN CULTURE WHERE WE HAVE CIRCUITRY FOR LATENCY BUT THE ARGUMENT WAS THIS IS SIMPLY A CULTURE PHENOMENON IN-VIVO DESPITE THE TEMPORAL PROBLEMS. WHAT IS REALLY GOING ON IS THE CELLS ARE SILENCING. AND LATENCY IS EPIPHENOMENON WHERE ITS CELLS ARE SILENCING AND THE VIRUS IS FOLLOWING THE CELLS ALONG AND BEING SILENCED AND THAT'S HOW YOU'RE GETTING LATENCY. SO WE DID THE EXPERIMENT WHICH WE SHOULD HAVE DONE TWO YEARS AGO, AND THE EXPERIMENT IS AS FOLLOWS. WE ARE TESTING WHETHER THESE CELLS RELAXING FROM ACTIVE TO RESTING MEMORY CAN PULL THE VIRUS ALONG WITH THEM. SO YOU TAKE A FULL-LENGTH VIRUS AND NOW YOU KNOCKED OUT ENVELOPE SO IT'S A SINGLE-ROUND VIRUS. IN THE READING FRAME YOU HAVE A SHORT-LIVED TWO HOUR HALF-LIFE SO SEE SILENCING OF THE VIRUS RAPIDLY WITHIN TWO HOURS. AND THE EXPERIMENT IS FROM PATIENTS OR HEALTHY DONORS, YOU TAKE CELLS, ACTIVATE CD4 POSITIVE T-CELLS AND HYPERACTIVATE THEM AND USE THIS FOR TWO DAYS AND THEN THEY CAN BE AFFECTED AND WHEN THEY ARE INFECTED, REMOVE THE ACTIVATION FEEDS STIMULI AND LET THE CELLS RELAX FROM ACTIVATED TO EASTING. LET THEM RELAX OVER THE COURSE OF THE WEEK OR TWO AND SINCE YOU REMOVED ENVELOPE, THE VIRUS DOESN'T KILL THE CELLS QUITE AS QUICKLY SO IT MADE IT HARDER ON OURSELVES BECAUSE WE ARE GIVING THE VIRUS AND CELLS MORE TIME SILENCE THE VIRUS WHEN THEY SHOULD BE KILLING THE CELLS WITHIN 48 HOURS. SO THIS IS FIRST TO SHOW YOU THAT THE CELLS ARE SILENCING. HERE ARE TWO ACTIVATION MARKERS. CD69 AND CD25. THIS IS DAYS POST REMOVAL OF THE BEAST. THE CELLS RELAXED TO PRE-ACTIVATION LEVELS. FOR CD25 IT TAKES LONGER. WHAT HAPPENS TO THE VIRUS? VIRUS DOESN'T CARE. THE VIRUS CONTINUES TO ACTIVELY EXPRESS ITS GENOME. SO THERE ARE FOUR DATA POINTS HERE. TWO DIFFERENT PATIENTS TWO. DIFFERENT REPLICANTS. EACH ONE OF THESE IS FOUR DATA POINTS. SO THE VIRUS DOESN'T CARE. EACH THOUGH THE CELLS ARE SILENCING, THE VIRUS IS RESILIENT TO THAT SILENCING AND IT MAKES SENSE BECAUSE IF YOU HAVE A STRONG POSITIVE FEEDBACK LOOP, IT'S DIFFICULT TO SHUT IT DOWN BY EXTERNAL FACTORS T HAS TO BE SHUT DOWN INTERNALLY. SO THIS IS WHERE WE STOOD IF THE VIRUS IS NOT BEING SHUT DOWN BY CELLULAR RELAXATION, WHAT IS SHUTTING IT DOWN? WE HAD A MODEL IN THESE PAPERS AS WELL. I'LL SHOW YOU THE FIGURE FROM THE REVIEW PAPER THAT DESCRIBES ALL THIS THE SIMPLEST MODEL IS JUST LTR AND TAT POSITIVE FEEDBACK LOOP AND WE KNOW THE PROMOTOR TOGGLES BETWEEN ON AND OFF STATE. HERE IS THE POSITIVE FEEDBACK LOOP. WE MEASURED THESE AT THOUSANDS OF GENOMIC LOCI. WE HAVE ALSO MEASURED THESE TRANSCRIPTIONAL RATES AND TRANSLATION RATES AND THE DECAY RATES. WE HAVE A WELL PROGRAM TIDES MODEL WE CAN COMPUTATIONALLY RECAPITULATE USING STIMULATIONS FOR THE AFFICIONADOS IN THE AUDIENCE AND HERE IS WHAT THEY LOOK LIKE. SO TAT MOLECULES ON THIS AXIS AND EACH TRAJECTORY HERE WILL BE AN INDIVIDUAL CELL. SO CELLS TURN ON AND THERE ARE ENORMOUS FLUCTUATIONS IN TA. AND CHARACTERISTICS OF POSITIVE FEEDBACK LOOPS WHICH AMPLIFY FLUCTUATIONS. KEEP ON RUNNING THE SIMULATION FOR DIFFERENT CELLS. HUGE FLUCTUATIONS AND OCCASIONALLY WHEN YOU RUN INTO SIMULATIONS, THE FLUCTUATIONS ARE SO LARGE THEY CRASH INTO ZERO. AND THE PROMOTOR IS SO WEAK IT CAN'T RECOVER THIS ALSO LEFT WITH US TWO MAJOR QUESTIONS. ONE, THE VIRUS IS USING FLUCTUATIONS TO DRIVE A DECISION BETWEEN TWO DIFFERENT STATES, HOW DOES IT EVER COMMIT TO ACTIVE REPLICATION? BECAUSE IF CAN CONTINUALLY JUST GO EXTINCT IN TAT AND FALL OFF. IT WOULD CONTINUALLY BE DOING THAT. THE ACTIVE REPLICATION STATE WOULD BE VERY UNSTABLE. HOW DOES THE VIRUS COMMIT IF IT USES NOISE TO SELECT? GENERAL QUESTION IN FACT. AND THE SECOND QUESTION WAS, WE ARE SEEING THE VIRUS ENCODES LATENCY PROGRAM. WHY WOULD IT DO THAT EVOLUTIONARILY? SO I'LL ANSWER THOSE TWO QUESTIONS IN ONE SLIDE. SO FIRST THE COMMITMENT MECHANISM. THIS BOILS DOWN TO HOW TO YOU STABILIZE A NOISY CIRCUIT? IF YOU'RE USING NOISE FOR SELECTION SCHEME, HOW DO YOU ATTEN THAT IT NOISE ONCE YOU HAVE SELECTED? -- ATTENUATES. TRANSCRIPTIONAL AUTO REPRESSION WHERE THE PROTEIN FEEDS BACK AND REPRESSES PROMOTOR. FUNDAMENTAL LIMIT ON HOW MUCH NOISE CAN BE SUPPRESSED BY TRANSCRIPTIONAL AUTO REPRESSION AND NOT ENOUGH FOR THESE TYPES OF LARGE FLUCTUATION SYSTEMS. BUT THERE IS AN ALTERNATE SCHEME WHICH WE FOUND IN HIV WHERE IN THIS CASE, THE REV PROTEIN DEPLETES ITS OWN RNA. THAT'S AUTO DEPLETION OR WE CALL IT THAT. AND THIS BEATS THE FUNDAMENTAL ON NOISE SUPPRESSION. SO IN THIS SIMULATION, UNREG LATED SYSTEM, HERE IS THE TRANSCRIPTIONAL AUTO REPRESSION IN RED AND IN BLUE IS THE AUTO DEPLETION. YOU CAN SEE IT BEATS THE LIMITS ON NOISE SUPPRESSION AND THIS IS CURRENTLY IN PRESS. THE SECOND QUESTION, WHY WOULD THE VIRUS DO THIS? IN IS THE REASON THAT HYPOTHESES OF THE EPIPHENOMENON WHERE THE CELLS WERE RELAXING AND THE VIRUS IS FOLLOWING, THIS IS THE REASON THE HYPOTHESIS WAS FORWARDED BECAUSE WE DIDN'T UNDERSTAN WHY THE VIRUS WOULD ENCODE LATENCY. THE CANONICAL IDEA IS IF YOU HAVE A VIRUS WHICH NEVER GOESALATED ENT, IT WILL ONLY REPLICATE FITNESS OF THE VIRUS. WHEREAS LATENCY CAPABLE VIRUS WILL TURN ON SOMETIMES AND TURN OFF SOME TIMES AND WILL HAVE A DETRIMENT BECAUSE IT'S NOT ALWAYS REPLICATING. SO THIS VIRUS WILL WIN COMMENT ONE SITUATION. WE KNOW THESE VIRUSES, LENTIVIRUSES WHICH INFECT PRIMATES, ON THE AFRICAN CONTINENT, THEY TARGET MUCOSAL MEMBRANES WHICH HAVE VERY FEW ACTIVE CD4 T-CELLS. SO THIS VIRUS INFECTS ALL TARGET CELLS IN THE MUCOSA AND DESTROYS THEM AND NO WAY TO GET FROM THE TARGET FAR TARGET RICH ENVIRONMENT. IT'S EXTINCTION EVENT FOR THE VIRUS. WHEREAS, THE LATENCY CAPABLE VIRUS BY SOME PROBABILITY CAN TURN OFF AND CAN WAIT IT OUT AND MAKE IT TO A TARGET-RICH ENVIRONMENT BIKE MUCOSA, OR LIKE THE GUT OR LIMP. AND RECEDE INFECTION. SO THIS IS THE HYPOTHESIS. THIS WAS A MODEL WE FORWARDED. IT FIT ALL THE AVAILABLE DATA. CALCULATE IT OPTIMAL LATENCY FREQUENCY AND IT FITS IN CULTURED LATENCY FREQUENCIES YOU OBTAINED AND IN PATIENTS WHICH IS STILL A MODEL WE HAVE TO TEST BUT THIS IS WHERE WE ARE RIGHT NOW AND I THINK SINCE RAVI IS STANDING HERE, I'M GOING TO SKIP THE SECOND PART AND NOT TELL YOU HOW TO MA NIP NOISE BUT THAT WAS PUBLISHED IN 2014. -- MANIPULATE NOISE. THANK YOU VERY MUCH. [ APPLAUSE ] >> ONE QUESTION. >> CAN I ASK TWO QUESTIONS? ONLY ONE QUESTION? OKAY. I WAS WONDERING WHEN YOU EXPRESSED THE TAT IN TRANS, ARE THE FLUCTUATIONS GOING AWAY OR DO YOU STILL SEE THEM? >> GOOD QUESTION. EXPRESSED IN TRANS, AND THE FLUCTUATION IS DEPENDENT ON THE PROMOTOR YOU USE N THIS CASE, THE DOXYCYCLINE INDUCIBLE. IT'S A HIGHLY-PRESSED PROMOTOR. YOU STILL HAVE FLUCTUATIONS BUT NOT INSIDE A POSITIVE FEEDBACK LOOK ANYMORE. >> BUT TRANSCRIPTION SITE FROM LT SYMPT ALWAYS THE SAME WHEN YOU HAVE -- HAVE YOU MAPPED LTI TRANSCRIPTION SITE? >> THE NSCRIPTION WE MAPPED IT. MANY PEOPLE MAPPED THOSE SITES. >> SO NO CHANGES? >> WHEN YOU'RE SUPPLYING IN TRANS? >> NO, JUST IN THE REGULAR NOISY. >> THERE IS NO CHANGES IN INITIATION SITE. AND YOU CAN SEE THAT IN A NUMBER OF DIFFERENT WAYS. >> THANK YOU. >> GOOD QUESTION. >> THANK YOU. [ APPLAUSE ] >> OUR NEXT SPEAKER IS DANIELA WHIT EN WHO WILL TALK ABOUT INTRODUCING THE DIMENSION OF TIME INTO MACHINE LEARNING AND GRAPH THEORETICAL APPROACHES. >> THANK YOU. TODAY I'M GOING TO TALK ABOUT MY WORK AS AN EARLY INDEPENDENCE AWARDEE LEARNING THE STRUCTURE OF A GRAPHICAL MODEL. AND SO GRAPHICAL MODELING HAS BEEN A REALLY POPULAR TOPIC IN THE STATISTICAL MACHINE LEARNING COMMUNITY IN RECENT YEARS. AND PART OF THE REASON THEY HAVE GAINED A LOT OF POP LATERY IS BECAUSE THEY GIVE US AN EFFICIENT WAY TO REPRESENT COMPLEX RELATIONSHIPS AMONG A REALLY LARGE NUMBER OF RANDOM VARIABLES. AND GRAPHICAL MODELS HAVE BEEN USED IN A LOT OF FIELDS BUT IN PARTICULAR IN BIOLOGY THEY ARE REALLY USEFUL AS INVESTIGATORS ARE INCREASINGLY COLLECTING REALLY LARGE DATASETS AND THEN THEY WANT TO A WAY TO REPRESENT THE DATE ON TO FIGURE OUT WHAT IT MEANS. SO, HERE IS AN EXAMPLE OF A GRAPHICAL MODEL WE SEE A LOT IN BIOLOGY. SO THIS IS AN EXAMPLE OF A GENE REGULATORY NETWORK. IN THIS FIGURE, EACH OF THE LITTLE CIRCLES REPRESENTS A GENE THE. SOME OF THE CONNECTED BYLINES, WHICH ARE EDGES. AND THE EDGES REPRESENT SOME KIND OF DIRECT RELATIONSHIP BETWEEN THE GENES THAT THEY ARE CONNECTING. THE EXACT MEDIA OF THAT RELATIONSHIP IS CONTACT SPECIFIC AND DEPENDS ON WHAT KIND OF STATISTICAL MODEL WE ARE USING IN ORDER TO MODEL THIS GRAPH. BUT THE IDEA SHEAR THAT SOME INVESTIGATORS COLLECTS A BUNCH OF GENE EXPRESSION DATA FIR A NUMBER OF OBSERVATIONS SO MEASUREMENTS FOR A LOT OF DIFFERENT GENES AND THE GOAL IS TO GET A PICTURE LIKE THIS ONE THAT HELPS US UNDERSTAND WHAT ARE THE RELATIONSHIPS AMONG THE GENES THAT HAVE BEEN MEASURED? SO ANOTHER CONTEXT IN WHICH WE SEE GRAPHICAL MODELS IN BIOLOGY IS WITHIN THE AREA OF FUNCTIONAL CONNECTIVITY MODELING. SO AN INVESTIGATOR GOES OUT, RUNS A BUNCH OF PATIENTS THROUGH AN FMRI MACHINE, AND THEN THE GOAL IS TO UNDERSTAND WHICH REGIONS OF THE BRAINS ARE ACTIVATING BUT IN PARTICULAR WHERE, WHICH REGIONS OF THE BRAIN ARE COACTIVATING TOGETHER. SO, IN THIS GRAPH, WHICH IS HARD TO SEE IN THIS SUPERIMPOSED ON TOP OF A MODEL OF THE BRAIN, WE HAVE BLUE CIRCLES WHICH REPRESENTS REGIONS OF INTEREST THAT WERE ASSAYED AND THEN SOME OF THESE REGIONS ARE CONNECTED BY A LINE AND THIS REPRESENTS TWO REGIONS OF INTEREST WHICH BASED ON OUR ALGORITHMS, WE THINK ARE CO-OCCURRING AND CO-ACTIVATING. SO, A LITTLE BIT MORE FORMALLY WHAT DOES THIS GRAPH REPRESENT? A EXAMPLE OF A GRAPHICAL MOD WEL9 NODES SO EACH CIRCLE IS CALLED A NODE. -- MODELS AND NODES. SO IF WE ARE LOOKING AT GENE EXPRESSION DATA OUR GRAPH WOULD HAVE 20,000 NODES AND IT WOULD BE A LARGE GRAPH THINKING ABOUT FUNCTIONAL ACTIVITY MODELING. SOME ARE CONNECTED BY EDGES. SO WE CAN HAVE A DIRECTED GRAPH LIKE THIS WHERE THE EDGES ARE ARROWS OR UNDIRECTED GRAPH IN WHICH THEY ARE JUST LINES. BUT IN THE CASE OF A DIRECTED GRAPH LIKE THIS, THIS EDGE MEANS THAT IN SOME WAY, THE SECOND NODE IS DIRECTLY EFFECTING THE FIRST NOSED. FOR EXAMPLE, MAYBE A CHANGE IN EXPRESSION AND THE SECOND NODE CAUSES A CHANGE OF EXPRESSION IN THE FIRST NODE. SO WHAT I'M GOING TO TALK ABOUT TODAY IS A REALLY CHALLENGING PROBLEM WHICH IS LEARNING THE STRUCTURE OF THE GRAPH. SO THE IDEA IS WE HAVE DATA. IT MIGHT LOOK LIKE THIS. AND OBSERVATIONS ON THE KEY VARIABLES WE MEASURED. SO MAYBE A FEW HUNDRED PATIENTS FOR WHOM WE MEASURED GENE EXPRESSION LEVELS IN THE COLUPS AND WE WANT TO TAKE THIS RAW DATA AND CONVERT INTO A PICTURE THAT LOOKS LIKE THIS WHERE THE CHALLENGE IS FIGURING OUT WHERE THOSE EDGES ARE. SO FIGURING OUT WHERE WE SHOULD BE PUTTING ARROWS. SO OF COURSE IN ORDER TO DO THIS, WE NEED A FORMAL STATISTICAL MODEL BECAUSE WE NEED TO FORMALIZE THE NOTION OF AN ARROW BUT FURTHERMORE WE NEED TO DO THIS IN A WAY THAT IS VERY OFFICIAL STA KISTICALLY BECAUSE IF YOU LOOK AT THE SHAPE OF THIS DATA MATRIX, WE HAVE MANY MORE COLUMNS AND MANY MORE VARIABLES WE MEASURED THAN OBSERVATIONS. WHICH MAKES IT VERY CHALLENGING TO DO ANY KIND OF STATISTICAL INFERENCE. SO THE MAIN CHALLENGE IS REALLY DIMENSIONALITY OF THE PROBLEM AND TO SEE THIS ON NOTICE AN EXAMPLE WITH 9 NODES WE HAVE 9 TIMES 8 OR 72 POTENTIAL EDGES. OUR GOAL IS TO FIGURE OUT OF THE POTENTIAL EDGES, WHICH ONES ARE TRUE EDGES. SO IN AN EXAMPLE OF 9 NODES, WE HAVE 72 POTENTIAL EDGES, AND IN AN EXAMPLE OF 20,000 NODES, CORRESPONDING TO GENE EXPRESSION DATA, WE HAVE ON THE ORDER OF HALF A BILLION POTENTIAL EDGES. SO THE DIMENSIONALITY OF THIS PROBLEM SCALES REALLY FAST AND AGAIN, THE PROBLEM IS THAT OUR RAW DATA IS WHAT IS SHOWN ON THE LEFT. WE TECHNICALLY HAVE NOT SO MANY OBSERVATIONS FROM WHICH WE WANT TO LEARN THE STRUCTURE OF THIS GRAPH. SO OUR GOAL IS TO DEVELOP AN ALGORYTHM THAT IS CORRECT UNDER REALISTIC ASSUMPTIONS. SO, WHEN I SAY CORRECT HERE, WHAT I SHEEN WE WANT TO DEVELOP A SET OF STATISTICAL MACHINE LEARNING ALGORITHMS THAT COMES WITH THEORETICAL GUARANTEES. IF YOU GIVE ME DATA I CAN DRAW A PICTURE AND TELL YOU THIS IS THE GRAPH THAT COMES WITH THE DATA BUT WE'RE NOT GOING TO KNOW IF I'M RIGHT OR NOT. THE PROBLEM IS LEARNING THE STRUCTURE OF A GRAPH IS A UNSUPERVISED PROBLEM IN MACHINE LEARNING. SO THAT IS WHAT YOU HAVE WHEN YOU HAVE A DATAET AND YOU'RE TRY TO GO LEARN SOMETHING ABOUT THE DATA AND YOU DON'T HAVE A WAY TO CHECK YOUR WORK. THIS IS IN CONTRAST TO SUPERVISED LEARNING WHERE YOU'RE DOING PREDICTIONS AND THEN YOU CAN CHECK YOUR WORK ON AN INDEPENDENT DATASET. SO WHEN WE ARE LEARNING THE STRUCTURE OF A GRAPH, THIS IS UNSUPERVISED LEARNING PROBLEM, THERE IS NO WAY TO CHECK THE ANSWER I GAVE YOU. YOU CAN CHECK LIKE A COUPLE OF THE EDGES EXPERIMENTALLY AND SEE IF THAT IS RIGHT BUT IN AN EXAMPLE LIKE GENE EXPRESSION DATA WHERE THE GRAPH HAS HALF A BILLION POTENTIAL EDGES, AT MOST YOU'LL MAKE A DROP IN THE BUCKET IN TERMS OF CHECKING MY WORK. IF WE ARE NOT GOING TO BE ABLE TO CHECK OUR WORK IN A THOROUGH WAY EXPERIMENTALLY, WHAT WE NEED ARE RIGOROUS STATISTICAL GUARANTEES THAT OUR ALGORITHMS WILL BE GIVING US THE CORRECT RESULTS UNDER SOME SET OF ASSUMPTIONS. IN PARTICULAR, WE WANT THESE ASSUMPTIONS TO BE REALISTIC BECAUSE IF I COME UP WITH THIS AND I TELL,IS 100% CORRECT UNDER ASSUMPTIONS THAT NEVER IN A MILLION YEARS APPLY TO YOUR DATA, IT'S NOT A VERY USEFUL GUARANTEE. SO WHAT I'LL BE TALKING ABOUT TODAY ARE SOME OF THE ASSUMPTIONS WERE BEING MADE IN THE MODELING LITERATURE IN THE WAY IN WHICH OUR RESEARCH GROUP MODIFIED THESE ASSUMPTIONS IN ORDER TO BE ABLE TO LEARN THE STRUCTURE OF THE GRAPH IN A MORE REALISTIC SETTING. SO, ONE VERY BIG PROBLEM WITH EXISTING METHODS FOR LEARNING THE STRUCTURE OF THE GRAPH IS THEY ASSUME INDEPENDENT EDGES. SO WHAT I MEAN BY THIS IS THE STANDARD WAY PEOPLE WERE LEARNING THE STRUCTURE OF A GRAPH IS BY PUTTING A PRIOR ON THE EDGES. AND THAT PRIOR WAS INDEPENDENT OVER EVERY EDGE. SO YOU CAN THINK OF THIS AS WHEN I'M LEARNING THE GRAPH, I'M PUTTING A PRIOR ON THE GRAPH STRUCTURE WHICH IS BASICALLY LIKE THROWING DARTS AT A DART BOARD. DART AND IF THE DART HITS THE BOARD THEN I GET AN EDGE AND IF NOT I DON'T. I DO THAT HALF A BILLION TIMES AND THAT GETS ME THE GRAPH. THERE IS A LOT OF STRUCTURE HERE. A LOT OF NODINGS ARE CORRECT CONNECTED TO OTHER NODES. THIS IS NOT WHAT YOU GET IF YOU THREW DARTAT A BOARD AT RANDOM. EXISTING METHODS ASSUME A PRIOR THAT HAS EVERY EDGE INDEPENDENT FROM EVERY OTHER EDGE. SO FUNDAMENTALLY IT'S AN AASSUMPTION BEING MADE THAT IS NOT MET BY BIOLOGICAL DATA AS A RESULT THE TYPES OF GRAPHS YOU ESTIMATE USING EXISTING METHODS DON'T LOOK REALISTIC AND THE STATISTICAL THEORETICAL GUARANTEES UNDERLYING METHODS ARE FOUNDED ON A SET OF ASSUMPTIONED NOT VALID BEING THAT EVERY EDGE IS JUST OCCURRING AT RANDOM. SO IN ORDER TO SOLVE THIS PROBLEM, WE DEVELOPED APPROACH FOR LEARNING STRUCTURE OF A GRAPH WHERE INSTEAD OF JUST ESTIMATING A GRAPH WHERE THE EDGES ARE INDEPENDENT WHICH IS WHAT EXISTING METHODS ARE DONE AND WHAT IS SHOWN HERE, WE ESTIMATE A GRAPH AWE AS A SUM OF INDEPENDENT EDGES PLUS A TERM INVOLVING HUB NODES. SO FOR EXAMPLE, IN THIS FIGURE ON THE RIGHT, NODES 3 AND 7 SERVE AS HUBS IN THE SENSE THEY ARE HIGHLY CONNECTED TO A LOT OF OTHER NODES. NODE 3 HAS THAT IN CONNECTIONS AND NODE 7 HAS 4 CONNECTIONS. AND SO WE MODEL THE GRAPH WE ARE TRYING TO ESTIMATE AS THE SUM OF THESE TWO TERMS, AND THIS ALLOWS US TO LEARN MORE REALISTIC STRUCTURE THAN IS POSSIBLE IF WE HAD ONLY ASSUMED THE PRESENCE OF INDEPENDENT EDGES. ONE OF THE THINGS THAT MAKES THE FORMULATION POWERFUL IS THAT IN THAT FIGURE ON THE RIGHT, WE DON'T NEED TO PRESPECIFY THE THIRD AND 7st ARE HUBS, WE FORMULATE THAT WITHIN THE OPTIMIZATION PROBLEMS WHERE WE ALLOW THE DATA TO TELL US IF ANY OF THE NODES ARE HUBS AND IF SO, WHICH NODES ARE. SO ANOTHER PROBLEM WITH EXISTING METHODS FOR LEARNING STRUCTURE OF A GRAPH IS THAT THEY DON'T ALLOW FOR HETEROGENOUS DATA. IT'S INCREASINGLY COMMON FOR PEOPLE TO COLLECT MULTIPLE TYPES OF DATA ON A SINGLE SET OF SAMPLES. FOR EXAMPLE GENE EXPRESSION DATA, CHIP SEQ DATA, DNA SEQUENCE DATA AND WE MIGHT WANT TO CONSTRUCT A GRAPH CONTAINING HETEROGENOUS DATA TYPES ALL TOGETHER. BUT FROM A STATISTICAL MODELING FRAMEWORK, THIS IS A PROBLEM. BECAUSE CHIP SEQ DATA TYPICALLY IS ACCOUNT VALUES VALUED AND YOU WANT TO MODEL IT AS A NEGATIVE BINOMIAL. SEQUENCE DATA IS BINARY SO WE WANT TO MODEL WITH A DISTRIBUTION. GENE EXPRESSION DATE DEPENDING ON HOW YOU COLLECT AND NORMALIZE IT, IT IS CONTINUOUS VALUED SO MAYBE YOU MODEL IT AS A -- DISTRIBUTION. BUT THE PROBLEM IS LIKE STEP 1, IN ANY ALGORITHM, WE NEED TO BE ABLE TO SPECIFY THE JOINT DISTRIBUTION OF THE RANDOM VARIABLES. WE NEED A SINGLE MODEL THAT DESCRIBES HOW ALL OF THE NODES IN THE GRAPH ARE ACTING TOGETHER. BUT IF MY RANDOM VARIABLES, IF THE NODES ARE OF DIFFERENT TYPES LIKE SOME ARE ACCOUNT VALUED, WE HAVE A PROBLEM WHICH IS WE CAN'T EVEN SPECIFY THE JOINT DISTRIBUTION. SO IN ORDER TO OVERCOME THIS, WE PROPOSED A FRAMEWORK THAT ALLOWS US TO MODEL THE NODE CONDITIONAL DISTRIBUTIONS INSTEAD OF THE JOINT DISTRIBUTION OF THE RANDOM VARIABLES. SO BASICALLY INSTEAD OF MODELING THE JOINT DISTRIBUTION, WE CAN MODEL THE CONDITIONAL DISTRIBUTION LIKE WE CAN THINK ABOUT THE DISTRIBUTION OF THE NODE CONDITIONAL ON ALL THE OTHER NODES AND CYCLE THROUGH THAT WAY. AND THIS IS JUST A VERY MINOR DIFFERENCE BECAUSE WE KNOW FROM BASIC STATISTICS THAT JOINT DISTRIBUTIONS AND CONDITIONAL DISTRIBUTIONS ARE CLOSELY RELATED BUT IT TURNS OUT IT IS EASY TO MODEL CONDITIONAL DISTRIBUTIONS WHEREAS JOINT DISTRIBUTIONS ARE HARDER. SO THROUGH THIS SMALL CHANGE, WE ARE ABLE TO APPLY THE GRAPHICAL MODELING FRAMEWORK TO DIVERT HETEROGENOUS DATASETS IN WHICH SPECIFYING JOINT DISTRIBUTION IS IMPOSSIBLE. SO ANOTHER PROBLEM WE ENCOUNTER SAID THAT EXISTING METHODS FOR LEARNING STRUCTURE OF A GRAPH CAN'T HANDLE TEMP ORAL DATA. I SHOWED YOU ON WHICH WE MIGHT WANT TO LEARN THE STRUCTURE OF A GRAPH AND MY OBSERVATIONS HERE IN THE ROWS OF THIS DATA MATRIX ARE INDEPENDENT. SO LIKE FOR EXAMPLE, YOUR ROAD COULD CORRESPOND TO DIFFERENT PEOPLE FOR WHOM YOU'RE DOING BRAIN IMAGING OR CELL LINES OR SOMETHING. SO VERY OFTEN INSTEAD, THE ROWS OF THE DATE MATRIX REPRESENT DIFFERENT TIME POINTS SO YOU HAVE ONE AND THE YOU'RE STUDYING IT OVER TIME. AND BASICALLY TWO PROBLEMS ARISE. ONE PROBLEM IS EXISTING METHODS FOR LEARNING STRUCTURE OF THE GRAPH ASSUME OBSERVATIONS ARE INDEPENDENT. FURTHERMORE, YOU REALLY MISSING OUT ON A LOT OF INFORMATION IN YOUR DATA IF YOU IGNORE THE FACT YOU HAVE ARE TIME POINTS FROM THE SAME SUBJECT IF YOU TREAT THOSE AT DIFFERENT SUBJECTS, YOU'RE MISSING THE OPPORTUNITY TO FOR EXAMPLE, INFER CAUSALITY ON THE BASIS THE DATA WAS COLLECTED OVER TIME. SO BASICALLY IN ORDER TO OVERCOME THIS PROBLEM, WE DEVELOPED A FRAMEWORK FOR GRAPHICAL MODELING FOR TWO TYPES OF TEMP ORAL DATA. WE CONSIDERED CONTINUOUS VALUE DATA AS ORDINARY DIFFERENTIAL EQUATIONS SO THIS IS USEFUL FOR EXAMPLE IF YOU ARE MEASURING GENE EXPRESSION OVER TIME IN ORDER TO MODEL A GENE REGULATORY NETWORK. AND WE ALSO CONSIDERED POINT PROCESS DATA LIKE YOU GET YOU'RE LOOKING AT A BUNCH OF NEURONS AND YOU WANT TO SEE HOW DOES ONE SPIKING EFFECT ANOTHER NEURON'S CHANCE OF SPIKING? SO WE MODEL THAT USING A CLASSICAL POINT PROCESS MODEL. SO THE LAST PROBLEM THEY WANT TO MENTION WITH EXISTING METHODS FOR LEARNING STRUCTURE OF A GRAPH IS A REALLY FUNDAMENTAL ONE IS SOMETHING PEOPLE DON'T LIKE TO TALK ABOUT IF THEY WORK IN THIS AREA BECAUSE IT IS A REALLY FUNDAMENTAL ASSUMPTION THAT IS LIKE NEVER GOING TO HOLD. AND IT SEEMS UNSURMOUNTABLE. SO THE PROBLEM IS THAT LATENT VARIABLES CAN SERVE AS CONFOUNDERS. WHAT DO I MEAN BY THIS? SUPPOSE THAT I'M INTERESTED IN STUDYING PEOPLE'S BRAIN ACTIVITIES. SO I SHOW A BUNCH OF STUDY PARTICIPANTS FROM SOME TELEVISION SHOW AND SCAN BRAINS WHILE WATCHING IT. AND MY GOAL IS TO PUT TOGETHER A PICTURE LIKE THIS SHOWING WHAT IS HAPPENING IN THEIR BRAIN WHILE WATCHING THIS SHOW. SO THE PROBLEM IS THAT METHODS FOR LEARNING THE STRUCTURE OF THE GRAPH, THEY ASSUME THIS IS A CLOSED SYSTEM, ASSUMING THAT EVERYTHING I NEED TO THINK, EVERYTHING THAT EFFECTS THE RANDOM VARIABLES HERE, WHICH IN THIS CASE ARE THE REGIONS OF INTEREST REPRESENTED IN BLUE, EVERYTHING HAS BEEN MEASURED AND NO OTHER VARIABLES THAT COULD BE RELEVANT THAT WEREN'T MEASURED. SO, WHAT DO I MEAN BY THIS? MAYBE IT TURNS OUT THAT THIS IS PROBABLY THE CASE MY BRAIN ACTIVITY DEPENDS HOW INTERESTED I AM IN WHAT I'M WATCHING, HOW DUDE MY ATTENTION SPAN IS, LOTS OF OTHER LATENT VARIABLES TOO NUMEROUS TO COUNT AND THESE ARE THINGS THAT AREN'T GOING TO BE PICKED UP DIRECTLY IN FMRI. SO THE PROBLEM IS, THAT THE PRESENCE OF THESE LATENT VARIABLES WHICH ARE SURELY PRESENT, IS GOING TO CAUSE FALSING IN AND I WAS FALSE POS I WAS IN GRAPH SPACE STATION. I'M GOING TO SEE ASSOCIATIONS BECAUSE I FAILED TO TAKE INTO ACT THESE LATENT VARIABLES AND FURTHERMORE I'M GOING MISS IMPORTANT ASSOCIATIONS AGAIN BECAUSE I IGNORED THESE VARIABLES. SO, IN A WAY, THIS SEEMS LIKE A REAL PROBLEM BECAUSE I CAN'T POSSIBLY MEASURE THESE. I MEAN, WE COULD THINK OF A REALLY LONG LIST OF POSSIBLE LATENT VARIABLES CONFOUNDING MY ANALYSIS. SO IN RECENT WORK WE ACTUALLY SHOWED THAT WITH CAREFUL ATTENTION TO EXPERIMENTAL DESIGN, WE CAN TRY TO OVERCOME THIS PROBLEM. AND SO IN PARTICULAR, TECHNICAL REPLICATES THAT IS LIKE IF YOU HAVE A SINGLE STUDY PARTICIPANT AND YOU ASSAY THEM TWICE INSTEAD OF ONCE, THEY HAVE SORT OF FALLEN OUT OF FAVOR IN A LOT OF AREAS OF BIOLOGY AND ALSO IN MACHINE LEARNING BUT WE SHOWED THAT TECHNICAL REPLICATES ARE REALLY USEFUL FOR THIS PROBLEM. BECAUSE IF INSTEAD OF JUST ASKING EACH OBSERVATION ONCE IF WE ASSAY IT TWICE, THEN WE CAN USE THE PRESENCE OF TECHNICAL REPLICATES IN ORDER TO CANCEL OUT THE EFFECT OF THE LATENT VARIABLES IN OUR DATA. SO THAT WE CAN INFER THE STRUCTURE OF THE GRAPH AS THOUGH WE HAVE SEEN THOSE LATENT VARIABLES WHICH WE DIDN'T SEE. SO I REALLY LIKE THIS RESULT BECAUSE AS A CLASSICALLY TRAINED STATISTICIAN, I OFTEN SEE EXPERIMENTAL DESIGN OF ISSUES FALLING BY THE WAYSIDE OF PEOPLE GETTING REALLY INTO BIG DATA AND MACHINE LEARNING AND THINGS LIKE THAT BUT THIS RESULT REALLY INDICATES THE FACT THAT YOU CAN NEVER STEP AWAY FROM THE IMPORTANCE OF EXPERIMENTAL DESIGN AND WELL-PLANNED STUDIES IF YOU WANT TO REALLY ANSWER THE BIOLOGICAL QUESTION OF INTEREST. SO I WANT TO SUMMARIZE ON WHAT I HAVE BEEN TALKING ABOUT FOR LEARNING THE STRUCTURE OF A GRAPH. THESE ARE CORRECT IN THE FACT THAT IN A SENSE THERE IS A LONG LIST OF MATHEMATICAL ASSUMPTIONS UNDER WHICH THIS IS GUARANTEED TO RECOVER THE STRUCTURE OF THE GRAPH IN AN APPROPRIATE SENSE AND OUR GOAL HAS BEEN TO MAKE ASSUMPTIONS AS REAL AS POSSIBLE. SO THAT WE CAN HOPE THAT IF WE APPLY THESE METHODS TO BIOLOGICAL DATA, WE ARE GETTING THE RESULTS THAT WE ARE LOOKING FOR. SO, THIS WORK WAS ALL DONE WITH A NUMBER OF COLLABORATORS AND THE THREE CONTRIBUTORS WERE THREE OF THE PH.D. STUDENTS FROM THE MY RESEARCH GROUP IN I DON'T YEARS AS WELL AS OTHER COLLABORATORS AT UDUB AND ELSEWHERE. I AM A BIG PRO OPPONENT IN PUBLICLY AVAILABLE SOFTWARE SO IT'S ALL AVAILABLE ON THE COMPREHENSIVE ARCHIVED NETWORK. AND I WANT TO THANK THE EARLY INDEPENDENCE AWARDS FOR THIS GREAT OPPORTUNITY AND ALL OF YOU FOR YOUR ATTENTION. THANK YOU. [ APPLAUSE ] >> THANK YOU FOR THAT TALK. I'M GOING TO FOCUS MY QUESTION ON ONE WORD, WHICH WAS THE WORD, CLEVER. TECHNICAL REPLICATES BY THEMSELVES AREN'T ENOUGH, THEY HAVE TO BE CLEVER AND I NEED TO LEARN MORE ABOUT MAKES A TECHNICAL REPLICATE CLEVER IN ORDER TO ADDRESS A LATENT VARIABLE. >> YOU'RE RIGHT. THE TECHNICAL REPLICATE, THERE IS NOTHING CLEVER THERE. I MEAN LIKE YOU'RE GOING TO STUDY 20 PATIENTS -- LET'S SAY YOU HAVE BUDGET TO RUN LIKE 20 SCANS, WHAT I'M TELLING YOU IS DON'T TAKE 20 DIFFERENT PEOPLE AND SCAN THEM. TAKE THEM ALL TWICE. SO THAT'S A STANDARD MEANING OF TECHNICAL REPLICATES. BUT THE THING THAT IS CLEVER IS WE ARE ABLE TO SHOW THAT WE CAN TAKE LIKE THE DATA FROM PERSON'S FIRST REPLICATE AND THE DATA FROM THAT PERSON'S SECOND REPLICATE AND CONTRACT OUT THEIR LIKELY FLOODS A CERTAIN WAY IN ORDER TO CANCEL OUT THE EFFECT OF THAT LATENT VARIABLE WE DIDN'T MEASURE. SO THE CLEVER THING WHAT WE DO WITH THOSE TECHNICAL REPLICATES AFTER WE COLLECT THEM. THIS IS SOMETHING THAT WOULDN'T BE POSSIBLE IF WE DIDN'T HAVE TECHNICAL REPLICATES. THIS IS EYE-OPENING BECAUSE IF A COLLABORATOR COMES TO ME AND SAY THEY ARE RUNNING A STUDY, THEY HAVE BUDGET FOR 20 SCANS, I WOULD JUST SAY GREAT G OUT AND FIND 20 PEOPLE AND RUN THEM. I WOULDN'T EVEN BEEN ABOUT IT BUT THIS WAS EYE-OPENING FOR ME BECAUSE WE SHOWED IT'S BETTER TO HAVE 10 PEOPLE TWO TIMES EACH THAN 20 PEOPLE ONCE EACH. >> VERY INTERESTING TALK. MY QUESTION IS ALSO ABOUT THE LATENT VARIABLES AND WONDERING IF YOU CAN CONTRAST THE INJURY APPROACH TO HAVE A RESTING STATE MODEL YOU'RE SUBTRACTING AWAY OR BASELINE OF SOME KIND. SO I HAVE SOMEONE MORE THANS EYE MOVEMENTS IN INANTS AND SO WE SHOW STIMULATIONS THAT WE CARE ABOUT AND IT'S MORE EQUAL TO GET THE TIME CONTINGENT RATHER THAN MEASURE THE SECOND TIME BECAUSE I HAPPEN TO BE LOOKING AT LEARNING OVER TIME. SO IF YOU HAVE THEM DO THE SAME EXACT TV VIEWING IN THE SCANNER A WEEK LATER, YOU'RE PROBABLY MEASURING SOMETHING ABOUT THE LEARNING SO THAT SUBTRACTION WON'T BE AS EFFECTIVE AS COMPARING TO A BASELINE IT'S TRICKY TO FIGURE OUT THE BASIS LINE. CAN YOU SAY SOMETHING ABOUT THAT AS OPPOSED TO RANDOM EFFECTS? >> THAT IS A GOOD QUESTION. SO HOW DOES THE BABY GET TWO TECHNICAL REPLICATES COMPARED TO THE IDEA OF GETTING A BASELINE AND THEN A MEASUREMENT USING THE SAME TASK? I MEAN IT'S POSSIBLE OUR APPROACH OF CANCELING OUT THE VARIABLES DUE TO TECHNICAL REPLICANTS COULD BE IN THE BASELINE F I'M HAVING SOMEONE DO SOME TASK OR WATCH A T.V. SHOW OR SOMETHING AND THEY ALREADY SEEN IT, THEN THAT WOULD BE BORING FOR THEM SO SOMETHING WOULD BE CHANGING. MAYBE ANOTHER EXAMPLE WHERE WE CAN SEE THE USE OF THE LATENT VARIABLES IS LAKELETS LET'S SAY I'M DOING COLLECTING GENE EXPRESSION DATA FOR A BUNCH OF -- OR SOMETHING. MUST BE A BETTER EXAMPLE IS FOR A BUNCH OF LIVER SAMPLES TAKEN FROM BARBITURATES SOME DISEASE WE ARE COLLECTING GENE EXPRESSION SAMPLES. THE UNDERLYING GENETIC BASIS FOR THOSE PATIENTS WILL BE DIFFERENT THAT SERVES AS LATENT VARIABLES. ALL DNA SEQUENCES ARE DIFFERENT. THAT WILL COME INTO UCLA EFFECT OUR GENE EXPRESSION LEVELS. -- WILL COME INTO PLAY AND EFFECT -- THE GENETIC BASIS WILL BE THE SAME ACROSS THE TECHNICAL REPENTS AND SO WE CAN LITERALLY CANCEL IT OUT BECAUSE THE LATENT VARIABLES STAYED THE SAME ACROSS THE REPLICATES. THE REPLICATES. >> SO IN THE INTEREST OF TIME WE NEED TO END THE SESSION. LET'S THANK DANIELA AND ALL THE SPEAKERS FROM THIS MORNING. [ APPLAUSE ] >> OUR SECOND PRESENTATION IN THIS SESSION IS FROM JAMES FRASER, I THINK WE'LL AGAIN SEE SOME OF THE IMPORTANCE OF DYNAMICS AND HE'S GOING TO TELL US ABOUT THE IMPACT OF MUTATION ON THE FUNCTION CONFIRMATIONS AND RECOGNITION OF UBIQUITIN. >> THANK YOU VERY MUCH, IT'S A REAL PLEASURE TO BE HERE AND A PLEASURE TO FOLLOW SUCH A TOUR DEFORCE TALK ON VERY DIFFICULT MEMBRANE STRUCTURES. I'M GOING THE TALK STRUCTURAL BIOLOGY ON A PROTEIN MORE TRACTABLE, UBIQUITIN. AND THAT'S GOING TO ALLOW US TO REALLY DIVE INTO THE DYNAMICS OF THE PROTEIN, HOW IT AFFECTS RECOGNITION AND IN VIVO TO YEAST TO SEE HOW DIFFERENT MUTATIONS PERTURBING THE FUNCTION OF UBIQUITIN IN CELLS. MY LAB BROADLY IS INTERESTED IN PROTEIN ALLOSERIA, WE THINK ABOUT PROTEINS AS CONFIRMATIONAL EQUILIBRIA AS ILLUSTRATED BY THE BLUE AND YELLOW PROTEIN AND HOW SMALL MOLECULES BINDING OR MUTATIONS CAN SHIFT THESE CONFIRMATIONAL EQUILIBRIA AND THE FUNCTION OF THE PROTEIN IN THE CELL. SOMETIME IT IS PERTURBATION IS SMALL MOLECULES BUT TODAY I'M GOING TO BE TALKING ABOUT PERTURBATIONS OF MUTATION AND HOW WE CAN READ THAT OUT BY FIRST X-RAY CRYSTALLOGRAPHY AND DEEP SEQUENCING. AS I MENTIONED WE'RE STUDYING PROTEIN UBIQUITIN HERE, WELL KNOWN AS A PROTEIN MODIFICATION TO PROTEIN MODIFIES OTHER PROTEINS FAMOUS FOR TAGGING THEM, FOR DESTRUCTION BUT ALSO QUITE A DYNAMIC LITTLE PROTEIN. WHICH EXHIBITS MULTIPLE CONFIRMATIONS IMPORTANT FOR ITS FUNCTION. AND THE WHOLE LANDSCAPE OF UBIQUITIN INTERACTION PROTEIN IS QUITE COMPLEX. THERE'S A REAL WIDE ARRAY OF DEDICATED MA -- MACHINERY NECESSARY FOR TARGETING DIFFERENT SUBSTRATE PROTEINS WITH UBIQUITIN FOR VARIOUS ROLES INCLUDING AS I MENTIONED PROTEIN DEGRADATION. FROM OVER THE PAST FEW YEARS MANY GROUPS INCLUDING MINE AND DEB'S HAVE BEEN INTERESTED IN THIS IDEA OF GENERATING UBIQUITIN MUTANTS THAT ACT AS INHIBITORS OF DIFFERENT UBIQUITIN MACHINERY IN THE CELL. SO I'M SHOWING CRYSTAL STRUCTURES FROM TAUGHT DOG'S LAB, UBIQUITIN MUTANTS IN GRAY WITH UBIQUITIN -- SORRY DIFFERENT UBIQUITIN INTERACTION PROTEINS IN GRAY AND DIFFERENT UBIQUITIN MUTANTS IN GREEN, WHERE MUTATIONS INDICATED IN MAGENTA ENDOW THE UBIQUITIN VARIANTS WITH DIFFERENT PROPERTIES SUCH AS BEING ABLE TO INHIBIT THESE UBIQUITIN PROTEINS. SO A FEW YEARS AGO I STARTED WORKING WITH A FRIEND OF MY FROM GRAD SCHOOL JACOB CORN THEN AT GENENTECH, HE SINCE MOVED TO UC BERKELEY HON TARGETING DEUBIQUITIN ACES, THAT REMOVE UBIQUITIN FROM SUBSTRATE PROTEINS. WITH THIS TYPE OF STRATEGY WHERE WE MAKE MUTATIONS, IN THE YOU CAN BIC TIN ITSELF AND ACT AS INHIBITOR. SO THE IDEA BEHIND THIS IS SOME DEUBIQUITIN AISS LIKE THIS, THEY'RE INTERESTED IN USP 7 PREFER THIS LOOP, THE BETA 1, 2 LOOP IN THE DOWN CONFIRMATION. SO I WILL PLAY A MOVIE HERE YOU CAN WATCH THE LOOP MOVE FROM THIS UP CONFIRMATION TO DOWN CONFIRMATION. AND REALLY THE CENTRAL IDEA HERE IS THAT THE LANDSCAPE, THE ENERGY LANDSCAPE OF WILD TYPE UBIQUITIN IS BIASED SUCH THAT IT'S DOMINATED BY THIS NON-USP P BINDING CONFIRMATION. WHAT JACOB WANTED TO DO IS USE ROSETA COMPUTATIONAL PROTEIN DESIGN TO CHANGE THE LANDSCAPES THAT DOMINANTLY POPULATE THE DOWN CONFIRMATION AND ACT AS A BETTER INHIBITOR OF USP 7. SO THEY FIRST PERFORMED A BUNCH OF DESIGN CALCULATION, AND THEN SOME PHASE DISPLAY TO ENHANCE THE MUTANT AND WHAT I WILL TELL YOU NOW IS HOW IT WORKED SO TAKE ON FAITH THAT IT WORKS AND YOU CAN READ ABOUT THAT. I WILL TELL YOU ABOUT HOW THE DESIGN PROTEIN CHANGED ITS CONFIRMATION AND ITS FUNCTION. TO DO THAT WE TURN TO X-RAY CRYSTALLOGRAPHY WHICH TAKE ONE POINT HOME FROM MY TALK LIKE IT TO BE THIS THOUGH WE'RE USED TO THINKING ABOUT THE RESULTS OF CRYSTALLOGRAPHY WITH A SINGLE STRUCTURE, SINGLE SET OF ORDINANTS DOWNLOADED IN THE PTB, IT'S A RICH ENSEMBLE EXPERIMENT OF MILLIONS OF MOLECULES THAT ARE I RANGED IN THE CRYSTAL LATTICE. SO WE DID TWO THINGS A LITTLE DIFFERENTLY IN MY LAB TO TRY TO EXTRACT OUT THESE MULTIPLE CONFIRMATIONS FROM CRYSTALLOGRAPHY EXPERIMENT. FIRST WE DON'T CROWD POOL THE CRYSTALS SO NORMALLY TO PREVENT RADIATION DAMAGE AND THERE ARE MANY GOOD REASONS TO DO THIS, WE HAVE CRYSTALLOGRAPHERS COOL CRYSTALS DOWN, THAT HAS DOWN EFFECT OF REMOVING THE BREATHING MOTIONS OF THE PROTEIN UNDERGOES. THE SECOND IS WE'RE GOING TO MODEL THE RESULTING ELECTRON DENSITY WITH MULTIPLE CONFIRMATIONS AND GET ATOMIC DETAIL OF THE DIFFERENT STATES OF THE PROTEIN ADOPT IN THE CRYSTAL. SO THAT PROCESS LOOKS LIKE THIS WHERE WE HAVE AN ELECTRON DENSITY MAP IN THIS SORT OF CLOUD REPRESENTATION HERE. WHAT MY LAB SPECIALIZED IN OVER THE PAST FEW YEARS IS IDENTIFYING LOW OCCUPANCY FEATURES THAT WEAK SIGNALS IN THESE ELECTRON DENSITY MAPS THAT REPRESENT THE DYNAMICALLY ACCESS CONFIRMATIONS OF THE PROTEIN. SO WE CAN SEE THAT THERE'S REALLY AN ADDITIONAL CONFIRMATION OF THIS LOOP HERE AND CURRENT FOCUS OF THE LAB IS REALLY IN FIGURING HOW TO AUTOMATICALLY FIT THESE TYPES OF FLEXIBLE BACKBONES AND LIGANDS THAT JUMP BETWEEN THESE RELATIVELY LOW ENERGY CONFIRMATIONS. WITH THIS RESULT IN IS MODELS WITH MULTIPLE CONFIRMATIONS AND ASSOCIATE OCCUPANCIES. THIS IS GIVING ATOMIC DETAIL INTO THE ACCESS CONFIRMATIONS NECESSARY FOR PROTEIN FUNCTION IN THE CELL. SO BACK TO THE UBIQUITIN MUTANTS. WE HAVE THE BEAUTIFUL DEFRACTION DATA FOR UBIQUITIN MUTANT AND MULTIPLE BACKBONE CONFIRMATIONS, YOU CAN SEE THE TRACE OF BETA 1, 2 LOOP IS MULTIPLE DIFFERENT CONFIRMATIONS FOR THIS INITIAL DESIGN MUTANT. WHEN WE CRYSTALLIZE AFTERPHAGE DISPLAY WE FOUND SINGLE BACKBONE CONFIRMATION TO INTERPRET WHAT THE ROSETTA DESIGN PROWAS DOING, IT WASN'T TILTING THE LANDSCAPE IN A WAY ORIGINALLY INTENDED. IN FACT WHAT IT DID IS CREATE A DYNAMIC PROMISCUOUS INTERMEDIATE MUTANT WHICH THEN THE PHAGE DISPLAY INHIBITED USP 7. SO THAT GIVES YOU A FLAVOR OF THE PROTEIN ENGINEERING AND CRYSTAL ORGANIZE WORK THAT WE DO. AND ONE OF THE SIGNATURES OF THAT WORK WAS REALLY THE GENERATING THE INHIBITOR REQUIRED MULTIPLE MUTATIONS. WE'RE ALSO REALLY INTERESTED IN WHAT HAPPENS TO THE CONFIRMATIONS AND FUNCTIONS OF UBIQUITIN WHEN YOU HAVE PERTURB WITH SINGLE MUTATIONS GIVEN UBIQUITIN IS EXTREMELY CONSERVED. THERE'S ONLY THREE SUBSTITUTIONS FROM YEAST TO HUMANS. WE PERFORMED A DEMUTATIONAL SCANNING IN ALL POINTS IN YEAST AND WE GROW A LIBRARY OF UBIQUITIN VARIANT, ALL SINGLE POINT MUTANTS. WE'RE ABLE TO COVER THAT WITH A WILD TYPE COPY WE CAN REPRESS AND OVER TIME THE FUNCTION OF THE PROTEIN OF THE UBIQUITIN MUTANTS IS WHAT DOMINATES THE FITNESS OF THE STRAINS AND OVER TIME WE GET THE FITTER STRAINS GOING MORE AND MORE AND WE CAN SIMPLY COUNT THESE UP WITH A DEEP MUTATIONAL SCANNING EXPERIMENT USING NEXT GENERATION SEQUENCING. SO THE WAY THAT WORKS IS WE HAVE LIGATED A BAR CODE BEHIND EACH MUTANT, WE THEN DEEP SEQUENCE IT AND WE GET A COMPLICATED MAP THAT LOOKS LIKE THIS. THE KEY POINTS OF THIS MAP IS THAT RED IS THICKER THAN WILD TYPE, BLUE IS THINNER THAN WILD TYPE AND THE WHITE WHICH REALLY DOMINATES THE MAP HERE, IS ESSENTIALLY WILD TYPE FITNESS. SO WHAT WE'RE LOOKING AT HERE IS A GRID OF ALL POSSIBLE MUTANTS ACROSS ALL POSSIBLE POSITIONS AND WE SEE THAT'S REALLY DOMINATED BY THIS WILD TYPE FITNESS. THIS LED US TO THIS QUESTIONS, WHY IS UBIQUITIN SO CONSERVED IN EVOLUTION? BASICALLY LOCKED IN TO ONE SEQUENCE. YET IN DEEP MUTATIONAL SCANNING EXPERIMENT IT CAN MUTATE ANY AMINO ACID. WE POSE THIS QUESTION KIND OF IN A UNIQUE WAY TO BUNCH OF DCF GRADUATE STUDENTS IN A PROJECT-BASED COURSE SO WE WANTED TO ASK THIS QUESTION IN A HIGH THROUGH PUT WAY, WHAT BETTER WAY THAN HIGH THROUGH PUT HARNESSING THE ENERGY OF 25 STUDENTS A YEAR IN A PROJECT BASED CLASS. SO WHAT THEY DO IS PERFORM THIS DEEP MUTATIONAL SCANNING EXPERIMENT IN DIFFERENT ENVIRONMENTAL CONDITIONS AND USING DIFFERENT CHEMICAL PERTURBATIONS. WHAT THEY UNCOVERED IS A TAPESTRY OF PERTURBATIONS SO WE LOOK AT THE DIFFERENT HEAT MAP WHERE IS BLUE IS MORE FIT THAN CONTROL DATA AND RED IS THICKER THAN THE CONTROL DATA WE CAN SEE THERE'S WIDE ARRAY OF PERTURBATIONS SHOWING SIX OF 15 CHEMICALS THAT WE PERTURBED YEAST WITH OVER THE YEAR. THIS UNCOVERED NEW ROLES FOR LYSINE LINKAGE AND HAS A LOT MORE NON-WILD TYPE VALUES THAN CONTROL DATA SO IT'S UNCOVERING THE CONSTRAINTS THAT ARE ACTING TON UBIQUITIN SEQUENCE. JUST HIGHLIGHTING ONE EXAMPLE OF WHAT THAT'S TEACHING US. WE HAVE DONE A PRINCIPLE COMPONENT ANALYSIS HIGHLIGHTING A SHARED SENSITIZING RESPONSE AMONG ALL THE MUTANTS, THAT CLUSTER DOWN AT THE BOTTOM BUT TWO OF THE PERTURBATIONS DTT AND RAPAMYCIN HAVE OPPOSITE TRENDS IN THE FITNESS DATA. THAT'S EXEMPLIFIED BY THEIR ROLE IN K-63 LINKED DYE UBIQUITIN SO I'M SHOWING THE STRUCTURE OF K-63 DYE UBIQUITIN WITH DIFFERENT FITNESS SCORES PAINTED ON THE STRUCTURE. THE LYSINE RESIDUES ARE OPPOSITE COLORS FOR THESE TWO DIFFERENT PERTURBATIONS THEN THE SURFACES ACTUALLY HAVE THE OPPOSITE COLORS OF THE LYSINE LINKAGES SO THAT IS TELLING US THE SURFACES ARE PERTURBED IN THE OPPOSITE DIRECTION OF THE LICENSE RESIDUE ITSELF. INDICATING A REAL LINKAGE BETWEEN THE POLYUBIQUITIN ASSEMBLY AND THE SURFACES THAT ARE INVOLVED IN ITS RECOGNITION. THIS IS FURTHER HIGHLIGHTED BY SIMILAR DELTA FITNESS PATTERNS FOR RESIDUES THAT IMMEDIATELY SURROUND K-63. YOU CAN SEE HERE THE DOMINANT BLUE COLOR IN ONE SET OF CHEMICALS AND DOMINANT RED COLOR IN SECOND SET, HIGHLIGHTING HOW NEIGHBORING RESIDUES ARE NEED FORD THE POLYUBIQUITIN CHAINS AND HOW ROLES OF SPECIFIC AMINO ACIDS AT THESE POSITIONS ARE NEEDED FOR RECOGNITION AND ASSEMBLY. THESE -- THIS IS REALLY TAPPING INTO A BROADER THEME IN THE INTEGRATION OF SYSTEMS AND STRUCTURAL BIOLOGY WHICH IS EXAMINING DELTA DELTA G MEASUREMENTS OR COMPENSATORY MEASUREMENTS THAT CAN DEFINE RESTRAINTS FOR STRUCTURED CALCULATION. THIS IS WHERE WE'RE GOING WITH THIS WORK IN THE FUTURE. SO MANY MEMBRANE PROTEINS BEFORE THE ADVENT OF CRYOELECTRON MICROSCOPY OR HIGH RESOLUTION CRYSTAL STRUCTURES ARE SPECULATIVE MODELS MADE OF THEIR TOPOLOGIES BASED ON PATTERNS OF COMPENSATORY MUTATION. SO HERE IS AN EXAMPLE FROM LILY JAN'S GROUP THAT NAILED DOWN WHICH EOCs ARE AD SWAY SENT BY COMPENSATORY -- ADJACENT BY COMPENSATORY MUTATIONS. EVENTUALLY THERE'S EXPLOSION OF WORK ANALYZING LARGE SEQUENCE ALIGNMENTS AND LOOKING FOR PATTERNS OF CO-EVOLUTION IN THESE ALIGNMENTS. THIS IS WORK BY DEBRA MARKS AND DAVID BAKERS LAB PROVIDE A POWERFUL CONSTRAINT FOR ASSEMBLING PROTEIN STRUCTURES USING CO-EVOLUTION PATTERNS IN THE SAME WAY WE USE AN NOE AND MR STRUCTURE DETERMINATION. WORKING WITH ANDREA SHALY AT UCSF WE HAVE BEEN INVOLVED IN TAKING THE GENETIC EXPERIMENTS DESCRIBED FOR UBIQUITIN AS WELL AS RNA POLYMERASE AND OTHER SYSTEMS TO THEN CALCULATE DISTANCE RESTRAINTS FROM THE PATTERNS OF THESE DIFFERENTIAL FITNESS VALUES. AND BE REALLY HAPPY TO TALK TO YOU ABOUT THAT, A WHOLE OTHER TALK OVER THE COFFEE BREAK LATER TODAY. THIS IS LEADING US TO A HYPOTHESIS THAT WE'RE GOING TO TAKE FORWARD WHICH IS THAT PERFORMING THE DEEP MUTATIONAL SCANNING EXPERIMENTS WITH DIFFERENT CHEMICAL PERTURBATIONS DEFINE QUANTITATIVE STRUCTURE FUNCTION RELATIONSHIPS AND WE CAN USE THOSE STRUCTURE FUNCTION RELATIONSHIPS TO REALLY DETERMINE THE ARCHITECTURES OF LARGE MACRO MOLECULAR COMPLEXES THAT ARE INTRACTABLE TO CRYOELECTRON MICROSCOPY. WITH THAT, I WOULD LIKE TO THANK THE PEOPLE WHO DID THE WORK, I HAVE A GREAT LAB INCLUDING A REALLY WONDERFUL HIGH SCHOOL STUDENT, JENNIFER LEE WHO DRAWS THIS UPDATED CARTOON OF THE LAB THAT'S INFLATED SUMMER SIZE EVERY YEAR. AND THE PEOPLE WHO DID THE WORK ESPECIALLY THE DEEP MUTATIONAL SCANNING WORK REALLY LED BY DAVID A GRADUATE STUDENT IN THE LAB AND STRUCTURAL UBIQUITIN MOWTANTS LED BY JUSTIN BEAL AS WELL AS COLLABORATORS AND FUNDING AGENCIES FOR THEIR SUPPORT, I WOULD LIKE TO TAKE ANY QUESTIONS THAT YOU HAVE AND THANK ROBBIE AGAIN FOR ORGANIZING THIS WONDERFUL SYMPOSIUM. THANK YOU. [APPLAUSE] >> WHEN YOU GET MAJOR AND MINOR CONFIRMATION, YOU HAD LIKE 30%, 70%, HOW LOW CAN YOU GO WITH THESE BRIGHT SOURCES LESS THAN 5%, WHAT LIMITS THAT? >> IT'S NOT ACTUALLY THE BRIGHTNESS OF THE SOURCE THAT LIMITS THAT, IT'S MORE THE RESOLUTION AND REALLY NOT THE BRIGHTNESS THAT'S LIMITING RESOLUTION, IT'S THE ORDER IN MOST CASES. WE GO DOWN TO 5 OR 10% WITH TRADITIONAL APPROACHES. WHEN WE DO A PERTURBATION SERIES, FOR EXAMPLE USING DIFFERENT TEMPERATURES WE GO DO DIFFERENT TYPE OF REFINEMENT TAKING ADVANTAGE OF THE INFORMATION ACROSS DATA SETS AND THERE WE CAN PUSH DOWN LOWER POTENTIALLY TO ONE OR TWO PERCENT. >> ANY OTHER QUESTIONS? ALL RIGHT. THANK YOU. [APPLAUSE] >> OUR THIRD SPEAKER THIS AFTERNOON IS GOING TO AT LEAST INTRODUCE ANOTHER PLAYER IN THIS THAT WE HAVEN'T SEEN IN THESE POSTERS AND THAT'S WATER. AND SO I WOULD LIKE TO INTRODUCE SONGI HAN FROM THE UNIVERSITY OF CALIFORNIA SANTA BARBARA WHO WILL TALK TO US ABOUT A SIGNATURE OF AN AGGREGATION-PRONE CONFIRMATION OF TAU. >> THANK YOU VERY MUCH. GREAT PLEASURE TO BE HERE AND SHARE WITH YOU OUR RESEARCH PROGRESS WE HAVE MADE AND UNDERSTANDING AGGREGATION MECHANISMS OF FROM TEEN IN PARTICULAR TAU. I PROMISE I WILL GET TO WATER IN ABOUT FIVE OR SIX MINUTES INTO THE TALK. SO MANY PROTEINS ESPECIALLY INTRINSICALLY DISORDERED PROTEINS FOR FIBER AND MANY DISORDER PROTEINS ESPECIALLY RECENT YEARS FOUND TO BE RNA BINDING IN A CERTAIN CONDITION EVEN CONDENSED AND FACE INTO PROTEIN -- BUT IT IS FAIR TO SAY THAT EVEN THOUGH A LOT OF EXCITING OBSERVATIONS HAVE BEEN MADE, THIS IS VERY IMPORTANT FOR PATHOLOGICAL PROCESSES AND AS WELL AS FUNCTIONAL PROCESSES THE CONSENSUS MECHANISM PREDICTING WHICH AGGREGATE AND FORM DROPLETS IN WHY AND QUESTION INTERVENENING THE PROCESS FAR FROM UNDERSTOOD. SO IN ORDER TO TAKE A CARTOON VERSION OF AGGREGATION HERE, FROM -- FIBRILS, ONE NEEDS TO UNDERSTAND WHAT IS THE DRIVER OF AGGREGATION TO THEN BE ABLE TO RELY BELIE INTERVENE SLOW DOWN OR PREVENT HYDROLLIZATION. LET ME SHOW THE TYPE OF WORK WE WERE ABLE TO DO WITH MEASUREMENT NOVEL AND MEASUREMENT METHOD UNDER TAU PROTEIN. TAU PROTEIN IS AN IMPORTANT INTRINSICALLY DISORDERED PROTEIN, ITS PARTICULAR RELEVANT TO TALK ABOUT THAT BECAUSE IN THE CASE OF TAU THERE IS A DIRECT RELATIONSHIP BETWEEN THE PATHOLOGICAL STAGES OF TAU AND TAU FIBERS IN THE BRAIN. ONCE CEASE INTRACELLULAR DEPOSITS THAT'S BAD NEWS AND DEPENDING ON LOCATION AND SPECIAL LOCATION AND MORPHOLOGY THERE'S A LOT OF CLINICAL DATA THAT CAN FROM THERE MAKE -- DRAW CONNECTIONS TO SPECIFIC DISEASE PATTERNS. D ON OTHER END OF THE SPECTRUM IS ALSO KNOWN FROM A LOT OF CLINICAL STUDIES THAT AS A PROTEIN LEVEL DIFFERENT MUTATIONS, VARIATION, POST TRANSLATIONAL MODIFICATION ALSO CORRELATED DIRECTLY TO PATHOLOGICAL SPECIES. AND THE QUESTION OF HOW AT THE PROTEIN LEVEL AND THEN FIBER LEVEL WHAT THE CONNECTION IS A DIFFICULT QUESTION. I WOULD LIKE TO TAKE EXAMPLE TAU PROTEIN AND FOR NOW TAKE THE IN VITRO AGGREGATION THAT WILL INDUCE BY HEPARIN AND ASK ABOUT THE SEED AND DRIVER OF AGGREGATION. HEPARIN IS USED IN MANY STUDIES AND WE JUSTIFY THAT BECAUSE FORM BY HEPARIN AGGREGATION RESEMBLE IN PATHOLOGICAL CASES BUT ALSO IN THE IN VITRO CASE WHEN YOU TAKE IN VITRO AGGREGATION PROCESS, THAT IMPORTANT MECHANISTIC QUESTION THAT ARE DEBATED AND NOD UNDERSTOOD. ALLOW ME TO TAKE NAIVE VIEW TO MAKE THE CASE WHY MECHANISM MATTER AND I SAY SIMPLISTIC BECAUSE OF COURSE IN REALITY IS MECHANISM MIGHT BE MUCH MORE COMPLICATE AND MANY MORE PROCESSES THAT NEED TO BE TAKEN INTO ACCOUNT. LET'S TAKE ONE VIEW, SIMPLEST VIEW ONE CAN TAKE WITH NUCLEATION AND POLYMERIZATION MECHANISM, THAT'S A PREVALENT VIEW ON AGGREGATION. YOU HAVE A PROTEIN MINOR POPULATION FORM FIBRIL, THEY'RE RECRUITING MORE PROTEINS TO FORM FIBRILS AND AGGREGATION PROCEEDS. ONE CAN RECONCILE THAT WITH A LACTASE AND VIEW THAT IS THERE FOR SURE FOR EXAMPLE FOR A BETA AGGREGATION. I CAN CONTRAST TO ANOTHER MECHANISM BUT I CO-DIRECT TRANSFORMATION MECHANISM. AGGREGATION PROPENSITY IS ENGRANDLATED AT PROTEIN LEVEL OR CO-FACTOR HERE HEPARIN CO-FACTOR LATER AND HOW THIS IS WHAT ARE OTHER CO-FACTORS CAN BE CONSIDERED. THERE SOMETHING GOING ON IN FIRST EARLY STAGES OF AGGREGATION THAT MAKE AS PROTEIN DESTINED TO AGGREGATION, IF ONLY THE FORMATION, PATHWAY ARRANGE ITSELF IN A ARRANGEMENT AND MATURE DIRECTLY GROW. YOU CAN SEE HOW THIS ENTIRELY DIFFERENT MECHANISM HAS DIFFERENT CONSEQUENCES IN CASE OF NUCLEATION POLYMERIZATION MECHANISM YOU MIGHT RECEIVE QUEST THE FIBRILS AND PREVENT AGGRESSION BECAUSE NOSE THOSE ARE THE SEED OF THIS PROCESS. IN A DIRECT TRANSFORMATION MECHANISM THEY HAVE HAVE NO EFFECT ON THE PROCESS THE LEVEL OF THE PROTEIN IS DESTIN TO AGGREGATION. IN MANY NEURODEGENERATIVE DISEASE, THERE ARE THERAPEUTIC APPROACHES OUT THERE DESIGNED TO SEQUESTER FIBRIL AND THOSE ARE KNOWN NOT TO BE SUCCESSFUL. AND MAY THE MECHANISTIC QUESTION MIGHT LIE AT THE CORE OF THIS AMBIGUITY. IN ORDER TO ANSWER THIS QUESTION EVEN SIMPLE MODEL SYSTEM BECAUSE I'M TAKING A TAU PROTEIN WE SYNTHESIZED IN A LABORATORY CONDITION, WE NEED WE NEED MEASUREMENT METHODS. CONVENTIONALLY ONE MIGHT DO LIGHT SCATTERING OR STAINING WITH THT, THOSE ARE TYPICAL MEASUREMENT METHODS AND ONE CAN DRAWER O OBSERVE A KINETIC CURVE BUT WHAT WE WANT TO ACCESS IS CONFIRMATION, CONFIRMATIONAL CHANGES IN SOLUTIONS IN SITU, IMMOBILIZATION, THROUGH PROTEIN DYNAMIC, BETA SHEATHE SIGNATURE AS WELL AS WATER AND I WILL GET TO THAT IN ORDER TO OBTAIN SUBTLE SIGNATURE OF PROTEIN CONFIRMATION. THE WATER INTERACTING WITH THE PROTEIN GIVE A SENSITIVE VIEW ABOUT CONFIRMATIONAL STATE OF THE PROTEIN. BUT EQUALLY IMPORTANT IS THE ABILITY TO MEASURE POPULATIONS, BECAUSE YOU HAVE A HETEROGENEOUS MIXTURE. NOT ONLY DO WE NEED TO MEASURE CONFIRMATION AND INTERMEDIATE STATES BUT KNOW WHAT THE FRACTION OF THOSE SPECIES ARE. SO WHEN IT COMES TO MEASURING DYNAMICS AND STRUCTURAL CONFIRMATION AND POPULATION, MAGNETIC RESONANCE TECHNIQUES ARE UNIQUELY SUITED ESPECIALLY IF SENSITIVITY OF THIS TECHNIQUES CAN BE ENHANCED. ONE MIGHT USE ELECTRON SPIN TEXT OR SPIN PRO BASE TEXT TO LOOK AT THIS SPECIES, SO ONE CAN USE ELECTRON LABELING THAT CAN BE INTRODUCED PROTEIN TO MEASURE DISTANCES AND ONE CAN TRACK THOSE SPECTROSCOPY FROM THE SOLUTIONS ALL THE WAY TO FIBRILS. THOSE ARE PUSHING THE BOUNDARIES OF-SPECTROSCOPY BECAUSE YOU NEED HIGH LEVEL DELUSION IN ORDER TO TRACK INTRAPROTEIN DISTANCE CHANGES IN SITU AS AGGREGATION OF CURES. ONE CAN USE LINE SHAPE ANALYSIS OF THIS ELECTRON SPIN PROBES BASED ON PROPERTY ISOTROPIC ORIENTATION. DIFFERENTIATE MOBILE SPECIES IN CONTENT I WILL SHOW YOU IN AN EXAMPLE WHEN WE TALK ABILITY HOW AGGREGATION. -- TAU AGGREGATION. WE HAVE DEVELOPED IN MY LAB TECHNIQUE NUCLEAR POLARIZATION TO MAP OUT SURFACE WATER COUPLED TO THE PROTEIN WITH SITE SPECIFICITY. I WILL SHOW YOU THAT WHEN YOU LOOK AT THE WATER SIGNATURE, IT IS MORE SPECIFIC THAN SIDE CHAIN DYNAMICS. THE FIRST LINE OF DEFENSE WOULD BE HYDROGEN BOND WATER. WE HAVE DEVELOPED A MEASUREMENT APPROACH TO MAP THAT OUT. THE ADVANTAGE OF ELECTRON SPIN BASED PROBE TECHNIQUE IS LARGE SYSTEMS CAN BE STUDIED. HOWEVER I SHOULD EMPHASIZE THAT WHEN IT BECOMES POSSIBLE NMR BASED STUDIES DO OFFER UNIQUE ADVANTAGES YOU CAN OBTAIN PROTEIN CONFIRMATION WITH SPECIFICITY, WHAT RESIDUES ARE INVOLVED, MORE DETAILS ABOUT THE STRUCTURE, ANTI-PARALLEL IN -- (INDISCERNIBLE) VERY IMPORTANT MORPHOLOGIES THAT ARE RELEVANT FOR DIFFERENT DISEASE TYPES. AS WELL AS IDEALLY CLUSTER ON ONE SIZE, THOSE MEASUREMENTS ARE POSSIBLE. AND ALL OF THIS COMES DOWN TO THE CHALLENGE OF SINCE SENSITIVITY. EXCEPT FOR LINE SHAPE ANALYSIS MEASUREMENTS CONVENTIONAL APPROACHES AND SENSITIVITY IS THE BIGGEST CHALLENGE. IN THE CASE OF WATER MEASUREMENT AS I WILL SHOW YOU IN A BIT, DRAMATIC SIGNAL AM MR. IFFYCATION TECHNIQUES WERE NECESSARY IN ORDER TO BE ABLE TO TRACK THIS DIFFERENT INTERMEDIATE SPECIES BY NMR, A HE CAN THE MEEK CALLED NUCLEAR POLARIZATION WILL BE NECESSARY TO GET AT THAT LEVEL OF ANALYSIS. WHAT I WILL DO IS I WILL START WITH A QUESTION OF TAU AGGREGATION AND THEN AS WE TALK ABOUT THIS, MECHANISM, THEN USE OR INTRODUCE THIS TECHNIQUE AS NEEDED. LET'S FIRST START WITH THE QUESTION ABOUT INTERMEDIATE FORMATION. IS THIS A MAJOR EVENT? ONCE WE KNOW THAT THE QUESTION IS ARE THEY ON PATHWAY? WHAT WE DO FIRST WE USE LINE SHAPE ANALYSIS BECAUSE THAT ALLOWS US TO DIFFERENTIATE BETWEEN MOBILE AND IMMOBILE COPULATION. IN SITU. SO WE CAN OBTAIN THIS SPECTRA DECONVOLUTE BETWEEN MOBILE AND IMMOBILE SPECTRA, LET ME SHOW YOU THE RESULT FIRST AND I WILL SHOW YOU WHERE VALIDATION COMES IN. WHAT YOU SEE HERE IS GREEN, IMMOBILE POPULATION PERCENT AND BLUE IS THE SPECIES THAT ARE STILL FREELY MOVING. PART OF THE PROTEIN GETS PACKED INTO, AND THE 444 LLC TERMINAL OF THIS TAU PROTEIN. WHAT WE SEE IN FIVE MINUTES WHAT WAS IN THAT TIME IN FACT INITIATING AGGREGATION WITH HEPARIN MORE THAN 50% OF THE PROTEINS ARE IMMEDIATELY IMMOBILIZED. WHAT THIS IS CONFIDENCE THAT THIS MOBILE AND IMMOBILE FRACTION CORRESPOND TO PHYSICAL ENTITY AS ILLUSTRATED HERE. IS THAT WHEN WE MEASURE THE ROTATIONAL CORRELATION TIME OF THIS SPACE, IT CONVERGES TO THE SAME VALUE OVER AS FUNCTION OF TIME WHEN WE ANALYZE DIFFERENT SPECTRA. SO WE CAN TALK ABOUT A MOBILE SPECIES, AND IMMOBILE SPECIES. AND CAN LOOK AT THE POPULATION. SO WHAT WE CAN SAY FOR NOW IS INTERMEDIATE FORMATION IN EARLY STAGE AGGREGATION. A QUESTION THEN IS THIS RELEVANT AND ON PATHWAY. RIGHT? TO ANSWER THAT QUESTION, WE NEED TO LOOK AT BETA SHEATHE SIGNATURE NEED TO FOLLOW EVOLUTION AND POPULATION. THIS' AN EPR TECHNIQUE THAT CAN DO THAT BUT SPIN EXCHANGE, I WILL BRIEFLY GIVE YOU A HINT WHAT THIS TECHNIQUE IS ABOUT. WHEN SPIN LABELS INTO FIBERS AND FIBRILS WHETHER PARALLEL OR ANTI-PARALLEL IN REGISTER OR OUT OF REGISTER, ONCE THEY PACK THE THIN LABELS COME WITHIN FIVE TO EIGHT ANGSTROM PROXIMITY OF EACH OTHER SO OVERLAP BETWEEN ELECTRON SPIN GIVES RICE TO A CHARACTERISTIC SO CALLED SKIN EXCHANGE COMPONENT. THIS IS THIS GREEN EXPERIMENTAL DATA OF AN IDEAL PEPTIDE SYSTEM. WE CAN APPLY THAT TO OUR PROTEINS THAT IS SUBJECT TO AGGREGATION AS FUNCTION OF TIME AND DO ANALYSIS. WHAT YOU CAN SEE HERE NOW IN YELLOW IS OVERLAID, THIS BETA SHEATHE POPULATION ON TOP OF WHAT I HAVE SHOWN YOU BEFORE WITH THE MOBILE IMMOBILE COMPONENT. SO YOU SEE AS EXPECTED IN THE FIRST 5 TO 20 MINUTES 5 TAU BETA SHEATHES AND 24 HOURS AFTER FIBER IS COMPLETE YOU HAVE 70, 80% OF THE TAU SPECIES IN BETA SHEATHE ARRANGEMENT SO THAT DOESN'T ANSWER THE QUESTION, IS IT ON PATHWAY OR NOT? SO WHEN YOU ENHANCE THE TIME RESOLUTION AND WE ACHIEVE THAT BY DOING FAST SPECTROSCOPY, FAST IN THIS CASE IN TERMS OF SECONDS, WHAT YOU SEE IS THAT INITIALLY THE MAJORITY COMPONENT HERE ENTERFACIAL COMPONENT IN GREEN, DIRECTLY TRANSFORMS WITHIN A COUPLE OF HOURS JUST AS NUMERICCAL COMPARISON INTERSPATIAL SPECIES ARE TRANSFORM LD JUST BY LOOKING AT POPULATION INTO BETA SHEATHE SPECIES. HOWEVER, THIS IS NOT A SATISFACTORY ANSWER, WE WOULD LIKE TO SEE MOLECULAR LEVEL EVIDENCE THAT WE ARE HAVING -- WE ARE HAVING PRE-DISPOSED INTERMEDIATE SPECIES THAT ARE PREARRANGED DESTINED TO FORM FIBRILS. AND ESPECIALLY IF YOU WANT TO TRACK MINORITY POPULATION THIS BECOMES AN IMPORTANT ABILITY TO EXPERIMENTALLY ACCESS. SO NOW I COME TO WATER. THIS IDEA IS SOMETHING WE WORKED ON A LONG TIME. IN FACT THE INSTRUMENTATION UP AND ESTABLISH IN THIS METHODOLOGY OF LOOKING AT WATER TO SAY SOMETHING ABOUT PROTEIN CONFIRMATIONAL SIGNATURE IS SOMETHING WE HAVE TO DEVELOP WITH SUPPORT OF THE INNOVATOR GRANT. PRINCIPLE IS AS FOLLOWS WE HAVE ELECTRON SPIN IN FORM OF NITROXIDE LABEL PAIR OF ELECTRON IT IS SAME LABEL WE USE FOR DISTANCE MEASUREMENT. THEY COUPLE THE PROTON SPIN OF WATER, THIS MEASUREMENT THAT CARRIED OUT 8 TO 12 GIG HERTZ EPR FREQUENCY WHICH MAKES THIS MEASUREMENT OF COUPLING SENSITIVE INVERSE 8 TO 12 GIGAHERTZ IN ORDER, 10 TO HUNDREDS OF SECONDS OF WATER DYNAMICS. WHICH IS A DYNAMIC WHERE WATER REFLECTING ON THE HYDROGEN BOND NETWORK STRENGTH OF SURFACE WATER. FROM THE WAY THE MEASUREMENT TECHNIQUE WORK, I WILL SAY ELECTRON SPIN THEY COUPLE IN RIGHT DYNAMIC RANGE AND READ OUT LARGE SIGNAL ENHANCEMENT THAT MAKES TECHNIQUE NOT JUST SENSITIVE TO DYNAMICS BUT OVERALL SENSITIVE TECHNIQUE ALLOW US TO MEASURE WATER ON THE SURFACE OF PROTEIN. SO THROUGH THE YEARS WE HAVE LOOKED AT IN SYSTEM TO VALIDATE THIS TECHNIQUE AND KNOW THAT WE CAN DIFFERENTIATE SURFACE AND BURIED WATER. SO THAT'S THE FIRST STEP. BUT MORE RECENTLY WE WERE ABLE TO SHOW NOT JUST THAT, BUT TAKE THE SURFACE WATER WHEN THE SITE IS PART OF A TOP LOGICALLY RICH SURFACE OR INTRINSICALLY DISORDERED, THERE IS A CLEAR SIGNATURE AND EVEN TAKE A COMPLETE LIST EXPOSE SITE OF PROTEIN CONCAVE VERSUS CONVEX, CHARGE AND NOT CHARGE, SURFACE DYNAMIC IS DISTINCT FROM SIDE TO SIDE AND TELL US ABOUT LOCAL PROTEIN ENVIRONMENT. SO NOW LET'S APPLY THAT TO TAU AGGREGATION IN SOLUTION AT ROOM TEMPERATURE AND TRACK THOSE SURFACE WATER DYNAMIC SIGNATURE. WHAT YOU SEE HERE ON THE Y AXIS TRANSLATIONAL CORRELATION TIME, HIGH SLOW E LOWER AN FASTER. YOU SEE HOLES HERE SO ON INTRINSICALLY DISORDERED SAY FOR NOW POLYMER, ALL DYNAMICS LOOK SIMILAR NOT SURPRISINGLY. WHEN YOU ADD HEPARIN, THIS IS HOW WE INITIATE AGGREGATION WITHIN THAT TIME OF MIXING, IMMEDIATELY YOU SEE A JUMP AND A DISPERSION. WHEN LOOKING AT THE IMMOBILE MOBILE FRACTION THROUGH PROTEIN SIDE CHAIN DYNAMICS, THEY WERE OVERALL SIMILAR. BECAUSE ALL SITES THOUGH PACKED POSSIBLY DIFFERENTLY ARE MORE OR LESS SLOW DOWN IN THEIR TUMBLING. WHEN YOU LOOK AT THE WATER SIGNATURE YOU SEE A CLEAR DISPERSION AND THE SITE THAT LATER DO PACK IN THE FIBERS ARE ONES THAT SHOW ONES THAT SHOW PATTERNS. WHERE YOU KNOW THEY ARE PREDISPOSED TO FORM BETA SHEATHE SIGNATURE. WHEN YOU LOOK AT THE BALANCE SURFACE WATER SIMILARLY YOU HAVE THE C TERMINAL SITES AND THE BOUND WATER SHOWING CLEAR DIFFERENCES OF THE SITES THAT PACK INTO BETA SHEATHE SIGNATURES. SO WE CAN CONCLUDE SO WE CAN CONCLUDE FROM HERE, LET ME JUST MAYBE SHARE ONE MORE THING. ONE MORE THING I WANT TO SHARE IS WE THEN ASK A QUESTION WHAT ABOUT CONFIRMATION BECAUSE THAT WAS IN MY TITLE, I WANTED TO MENTION THAT BRIEFLY. WE TAKE REGIONS THAT WE KNOW ARE COMPLETELY EXTENDED FIBRIL AND FOLLOW THAT IN SITU. AS AGGREGATION IS INITIATED. AND THE SLIDES AT THE END BASICALLY TELLS US WITHIN THAT TIME OF INITIATING AGGREGATION, IMMEDIATELY YOU SEE A CONFIRMATIONAL CHANGE. SO THEN I THINK I SHOULD JUST GO TO THE CONCLUSION. AND SHOW THAT WE WERE ABLE TO UNAMBIGUOUSLY IDENTIFY THE DIRECT TRANSFORMATION PATHWAY IS THE METHOD WHICH TAU AGGREGATES. IT'S A PROCESS THAT BEST BY HEPARIN OR CONFIRMATIONAL STATE IN GREEN AND WE BELIEVE THE DIFFERENT MUTATIONS THAT ARE KNOWN TO BE THESE RELEVANT WILL HAVE DIFFERENTIATING SIGNATURE AT THE PROTEIN CONFIRMATIONAL LEVEL THAT WE CAN STUDY WITH THE TOOLS THAT WE HAVE DEVELOPED IN MY LAB. THANK YOU. [APPLAUSE] >> SO WE HAVE TIME FOR SEVERAL QUESTIONS. >> HI. SO SENSITIVITY ON SAY THE AMOUNT OF MOLECULES THAT ARE -- THAT HAVE LOCALIZED SURFACE VERSUS WATER, WHAT I'M GETTING AT IS IF YOU HAVE A FEW OF THEM THAT ARE NOT ON THE AGGREGATE, THAT'S JUST LIKE FLIPPING IN AND OUT OF THE AGGREGATE, HOW MUCH OF THAT CONTRIBUTES TO THE TOTAL SIGNAL? >> IN THE CASE OF THE WATER SIGNATURE, IT WILL MEASURE AVERAGE, IN THE CASE OF WATER, OUR POPULATION WILL BE ONLY AS GOOD AS PROTEINS PARTICIPATING IN THIS EVENT. IN THE CASE OF CONFIRMATION AND THE MOBILE IMMOBILE FRACTION, WE GET DISTINCT FREQUENCIES AND WE CAN TELL WHAT PART IS MOBILE AND IMMOBILE. IN OTHER WORDS WE HAVE TO USE THIS EPA SUPPOSE ROCKSCOPY TO DO POPULATION ANALYSIS -- SPECTROSCOPY TO DO POPULATION ANALYSIS AND DO WATER SPECTROSCOPY TO IDENTIFY THE PROTEIN SPECIES. >> IN THE EXPERIMENTS YOU SHOW A GRADUAL CHANGE IN HYDRATION LEVELS. CAN YOU SEE THAT IN THE DEAR SPECTROSCOPY WITH AGGREGATION? DOES IT START OUT RANDOMIZED AND -- SHOWING THIS FIGURE SNAPPING IN ORDER BUT CAN YOU SEE THAT GRADUALLY? >> DISTANCE CHANGES WHAT WAS SURPRISING IS ABOUT THAT TIME FIVE MINUTES, THE -- DISTANCE IS DIRECTLY SNAP INTO THE DISTANCE FOUND IN THE FIBERS AND NO LONGER CHANGE BUT LOOKS LIKE HEPARIN IS INDUCING A CONFIRMATIONAL CHANGE IMMEDIATE WITHIN THAT TIME WE CAN CAPTURE AND THEN SLOWLY ASSEMBLY PROCESS IS WATER CAPTURING. CAPTURING SLOWER EVENTS THAN CONFIRMATIONAL EXTENSION. >> THE END TERMINAL INTERACTIONS TOO, NOT THE INTRA, CAN YOU SEE ANYTHING WITH THE DEER IN BETWEEN MOLECULE? >> YOU HAVE TO USE A DIFFERENT LABELING SCHEME, IF YOU MIX THOSE TWO. >> THANK YOU. >> SO OUR FOURTH SPEAKER THIS AFTERNOON IS BRIAN PAEGEL, HE IS GOING TO INTRODUCE ANOTHER TOPIC IN STRUCTURE THAT THESE DAYS WE ALL RECOGNIZE THAT SIGNAL TRANSDUCTIONS HAS A LOT TO DO WITH POST TRANSLATIONAL MODIFICATIONS OF PROTEINS. AND THE SALT RATION AND STRUCTURE, THAT IS SOME PART OF WHAT BRIAN IS GOING TO TELL US ABOUT THIS AFTERNOON. SO WITH HIS INTERESTING TITLE FINCHES AND SEEDS, PROTEASES AND BEADS: EVOLUTION OF NEW PROTEASE TOOLS FOR HIGH-THROUGH PUT POST TRANSLATIONAL MAPPING. >> THANK YOU FOR COORDINATING THIS WONDERFUL SYMPOSIUM, SORRY FOR MY SING SONG TITLE THERE. THIS IS THE ULK 1 KINASE, THE MASTER REGULATION TO OF AUTOPHAGY IN CELLS. KINASE COORDINATES THESE VIA PHOSPHORYLATION OF PORE PROTEINS AND ITSELF. IT PHOSPHORYLATES ITSELF IN AT LEAST 16 DIFFERENT POSITIONS THAT ARE SHOWN HERE IN ORANGE. IN THE POST GENOME ERA, THIS NET WORK OF POST TRANSLATIONAL MODIFICATION, MAPPING OF SITES AND TYPES OF PTMs PROTEINS THAT REPRESENTS ONE GRAND CHALLENGE FOR CHEMICAL ANALYSIS TODAY. AND THE CENTERPIECE ANALYTICAL TECHNOLOGY FOR ADDRESSING THIS IS HIGH RESOLUTION TANDEM MASS SPECTROSCOPY. YOUR SAMPLE PROTEIN IS DIGESTED WITH PROTEASE, TRYPSIN ALMOST ALWAYS TO GENERATE PEPTIDES, I'M SHOWING AN EXAMPLE OF A PEPTIDE HERE. THESE PEPTIDES ARE CHROMATOGRAPHICALLY ISOLATE AND INFUSED INTO THE ELECTROIONIZATION SOURCE MASS SPECTROMETER. WHERE THE INSTRUMENT THEN MASSES EACH OF THESE PEPTIDES TO PARTS PER MILLION FOR PRECISION. AND THEN EACH OF THE MOST ABUNDANT PEPTIDES IN THE STREAM ARE THEN SUBJECTED TO A SECOND DIMENSION OF MASS SPEC TROUGH METRIC ANALYSIS PEPTIDE IS FRAGMENTED AT ONE POSITION, ONE OF ITS BONDS ALONG THE BACKBONE. THIS COULD BE ANY ONE OF THE MI BONDS, IT COULD BE AT THE FIRST, SECOND, THIRD, THE WHOLE SERIES OF IONS GENERATED IN SECOND DIMENSION AND B ION SERIES ON TOP ARE ALL POSSIBLE ION FRAGMENTS ON END TERMINUS FROM THE END TERMINAL SIDE AND SIMILARLY ALL THESE FRAGMENTATION EVENTS CREATE Y ION SERIESS FROM THE C TERMINUS. THIS GROUP OF IONS THEN COMPRISES THE MS 2 SPECTRUM FROM WHICH WE CAN GLEAN THE SEQUENCE OF THAT PEPTIDE. AMINO ACID SEQUENCE BY DATABASE MATCHING. EVERY ONCE IN A WHILE THERE'S A MASS DIFFERENCE THAT DOESN'T CORRESPOND TO THE 20 RESIDUE MASSES FOR EXAMPLE IN THIS PEPTIDE THE SIX SERINE AS ADDITIONAL 80 UNITS WHICH CORRESPONDS TO PHOSPHORYLATION. SO BACK TO ULK 1 HOW ALL THE SITES AND PHOSPHORYLATION WERE IDENTIFIED BUT EYE NOT SHOWING REGIONS NOT SEQUENCED BY MASS SPECTROMETRY, THERE'S REGIONS THAT LIE IN AMINO ACID SEQUENCE BASE WHERE THERE ARE NO TRIP TICK SITES OR VERY FEW, OR ACTUALLY THERE ARE TOO MANY. ANY CASE THESE REGIONS ARE NOT DETECTED SO WE DON'T KNOW WHAT'S GOING ON. IT CAN BE ACTUALLY QUITE IMPORTANT FOR EXAMPLE IN THIS MIDDLE REGION WHICH IS THE SERENE AND PROLINE RICH WHICH HAS POTENTIAL SITES FOR PHOSPHORYLATION, WE CAN'T SEE THEM. THE ANSWER ROLLING IN YOUR HEAD AND MINE, JUST GET ANOTHER PROTEASE, THERE ARE SEVERAL OUT THERE THAT CLEAVE DIFFERENT PLACES CREATE PEPTIDES FOR ANALYSIS AND WE'RE BACK IN BUSINESS. THE REALITY IS THAT THAT'S NOT THE CASE. AND THIS WAS THE INSPIRATION FOR THIS PROPOSAL INNOVATOR PROGRAM. REALLY EVERYONE USES TRYPSIN, IT IS A SERINE PROTEASE, HIGHLY ACTIVE AND SELECTIVE THAT CLEAVES C TERMINAL TO ARGININE AND LYSINE. THERE ARE MAYBE FIVE OTHER PROTEASES THAT ARE IN ROUTINE USE IN HIGH RESOLUTION MASS SPECTROMETRY, ONE IS TRYPSIN RELATIVE OF TRYPSIN AND MASS SPECTROMETRIESES IN THE ROOM ARE SHIFTING UNCOMFORTABLY IN THEIR SEATS BECAUSE CHAI MOW TRYPSIN IS PROMISE SKEW WOWS TO BE USEFUL FOR ROUTINE USE BUT CLEANS SIDE CHAINS AND SMATTERING OF THE OTHER SIDE CHAINS. I WILL SHOW YOU ANOTHER FAMILY MEMBER OF THE CHYMOTRYPSIN FAMILY NOT USED, AS A LAST THAT CLEAVES GLYCENE, ALANINE AND VALINE. AND WHERE OTHERS MAYBE PROTEASES, I SAW AN OPPORTUNITY FOR EVOLUTION TO CREATE NEW TOOLS FOR US BECAUSE THIS FAMILY EXHIBITS DIVERGENT PROTEOLYTIC SPECIFICITY YET HAS HIGHLY HOMOLOGOUS STRUCTURE AND SEQUENCE. SO THIS TELLS ME THIS IS GREAT STARTING POINT FOR EVOLUTION. WE NEED NEW BEAKS. QUESTION NUDE MUTATIONS THAT ALLOW ACCESS TO PROTEOLYTIC SPECIFICITY. IN ORDER TO DO THIS WE NEED TO CREATE A COLLECTION OF MUTANT PROTEASES WITH DIFFERENT SPECIFICITIES AND SELECT PROTEASES WE'RE INTERESTED IN AND AMPLIFY AND CHARACTERIZE THEM. AND THAT TURNS OUT MUCH EASIER SAID THAN DONE. FIVE YEARS INTO THIS PROJECT. TRYPSIN IS A SERINE PROTEASE WITH A CANONICAL CATALYTIC TRIAD THAT FORMS THE BUSINESS END OF THE ENZYME THAT CATALYZES THE PROTEIN PEPTIDE BOND CLEAVAGE AND POSITIVELY CHARGED SIDE CHAIN BY SPECIFICITY POCKET THAT HAS ASPARTATE IN THE BASE OF THE -- IN ITS BASE THAT ENGAGES THOSE POSITIVELY CHARGED SIDE CHAINS ARGININE AND LYSINE OR SO WE WERE TAUGHT IN -- THERE'S A CLASSIC PAPER IN 1992 THAT SAID THAT SHOWED FROM BILL RUDDER'S LAB THAT SHOWED IT'S A LITTLE MORE NUANCED THATTEN THAT. THERE'S TWO SURFACE EXPOSED LOOPS THAT BOUND ACTIVE SITE, EACH FOUR RESIDUES LARGE, THAT REALLY ARE THE -- THAT FORM THE BASIS OF DETERMINING THE SPECIFICITY OR FAMILY OF ENZYMES. SO THE CHALLENGE THEN, AND ACTUALLY EVIDENCE IN THE LITERATURE SHOWING IF YOU TRANSPLANTED THE LOOPS FROM CRIME TRYPSIN TO TRYPSIN AND MADE OTHER SINGLE CHANGES YOU CAN RECONSTITUTE CHAI MOW TRIP TICK ACTIVITY IN SCAFFOLD BUT BARRING MEANS OF GETTING AT THE VERY LARGE SPACE OF MUTATIONS, THAT ARE INVOKED WHEN YOU TAKE SAY FOUR RESIDUES AND MUTAGENIZE THEM, 160,000 AMINO ACID THERE'S NO WAY TO MOVE THROUGH -- THERE'S IN WAY TO MOVE THROUGH PHENOTYPE SPACE USING MICROPLATES WHICH ARE CONVENTIONAL MODE FOR PROTEIN ENGINEERING ANDS EVOLUTION. BUT IN 1998 THERE WAS A REVOLUTIONARY PAPER FROM TOPIC DESCRIBING ULTRA MINIATURIZATION OF PROTEIN EVOLUTION IN THE FORM OF WATER AND OIL DROPLETS IN EMULSION. AND HERE SINGLE MOLECULES FROM YOUR GENE LIBRARY ENCANCEL LATED IN THE PICO LITTER SCALE WATER OIL DROPLETS 10 MILLION CAN INHABIT A TUBE UNLIKE THE 96 WELL PLACE HERE WHICH ARE HONRY PRETTY QUICKLY. THERE'S SEVERAL TRICKS BUTTRICK IS HOW DO YOU MARK IN THE TINY MICROSCOPIC VOLUME YOURS GENES WITH DESIRED ACTIVITY? THIS TURNED OUT TO BE QUITE THE ODYSSEY OF TECHNOLOGY DEVELOPMENT IN OUR LABORATORY, I WILL SHOW YOU HOW WE SOLVED IT. WE RELY ON MICROBEADS, THESE ARE 2.8-MICRON MAGNETIC BEADS SUBJECTED TO ROUTINE SOLID BASE BY FUNCTIONALLIZE THEM WITH BOTH A DNA PRIMER FOR PCR, AND A ROTE MEAN 110 PROTEASE ACTIVITY BASED PROBE. THIS ACTIVITY BASED PROBE WORK AS FOLLOWS, IN THE PRESENCE OF PROTEASE THAT RECOGNIZES SIDE CHAIN NOTED HERE GENERICALLY AS R 1 WHEN IT RECOGNIZES THE RO1 SIDE CHAIN AND CLEAVES THE BOND ADJACENT, IT DECLENCHES THAT ROAD MEAN FLUROPHORE GENERATING A FLUORESCENT BEAD. THE WAY WE ACCESS OUR GENE HIGH LIBRARY WE TAKE SITE SATURATION MUTAGENESIS LIBRARY WHICH IS TRIVIAL TO CONSTRUCT AND TAKING THE LEAD FROM THE WORK OF BURT VOGELSTEEN IN 2003 WE USE SOLID BASED PCR TO POPULATE SURFACE OF THESE GENE OR PRIMER PROBE BEADS WITH MANY COPIES OF EACH MEMBER FROM THAT LIBRARY. THE WAY IT WORKS IS EACH DROPLET CONTAINS TACT DNTP IN PRIME, THERE'S ALSO THE OTHER PRIMER ON THE BEAD AND AFTER THERMAL CYCLING THIS IS HAPPENING IN ALL DROPLETS AFTER THERMAL CYCLING DROPLETS THAT HAD A BEAD AND AT LEAST ONE GENE FROM THE LIBRARY ARE TEMPLATED WITH ABOUT 10,000 COPIES OF GENES OF THE SINGLE GENE PER BEAD WHICH WE QUANTITATED BY PCR. WE CAN THEN HAIR THESE THOSE BEADS AND REDISTRIBUTE INTO DROPLETS OF EMULSION IN VITRO TRANSCRIPTION TRANSLATION. AND THE TRANSCRIPTION TRANSLATION REAGENT WILL THEN TAKE THE DNA TEMPLATES BOUND TO THE BEAD TRANSCRIBE INTO mRNA AND TRANSLATE INTO PROTEASES. AND IN THE INTEREST OF TIME I WON'T GO INTO DETAILS HOW WE ENGINEER THE SYSTEM TO WORK WITH TRYPSIN WHICH HAS ITS OWN PERSONALITY. IN THE PRESENCE OF PROTEASE MUTANT FROM THE LIBRARY THAT CAN CLEAVE THE BEAD THE PROBE IS FLUORESCENT INDIVIDUALLY IN ITS OWN DROPLET, THIS IS IN PARALLEL FOR MILLIONS OF BEADS. AFTER THE EXPERIMENT WE CAN HARVEST THE BEADS SOME WHICH ARE FLUORESCENT, SOME ARE NOT, SOME WHICH HAVE GENES SOME WIDOW NOT BUT FLUORESCENT SHOULD BE THE ONES THAT ONLY HAVE GENES ON THEM AND SEPARATE THEM BY FLUORESCENCE ACTIVATED CELL SORTING. GOING BACK TO ULK 1 EXAMPLE, WE'RE LOOKING AT THIS AND GOING WE HAVE THIS PLATFORM FOR EVOLVING PROTEASES WITH ANY KIND OF SPECIFICITY, WHAT MIGHT WE GO AFTER? WE THOUGHT IT WOULD BE QUITE INTERESTING TO GO AFTER POST TRANSLATIONAL MODIFICATIONS THEMSELVES AS A TARGET PROTEOLYTIC ACTIVITY BECAUSE CLEAVAGE AT THE POST TRANSLATIONAL MODIFICATIONS WOULD UNAMBIGUOUSLY MARK THOSE AS HAVING BEEN MODIFY AND EASY TO FIND IN THE MASS SPECTRA DATA. IN FACT, THAT HAD BEEN SUGGESTED OVER A DECADE EARLIER IN THE LITERATURE FROM (INDISCERNIBLE) LAB. SO THE MODIFICATION WE ENDED UP GOING AFTER FOR A VARIETY OF REASONS, WAS SI TRILL THRENE. SY TRILLIZATION IS CATALYZED BY PROTEIN ARGININE DEAMINASES AND CATALYZE THE HYDROLYTIC DEAM NATION OF THE GROUP ON ARGININE SIDE CHAINS TO THE URETER GROUP HERE WHICH RESULTS IN CHARGE NEUTRALIZATION OF ARGININE SIDE CHAIN. AND PADS ARE VARIOUSLY INVOLVED IN A VARIETY OF IMMUNOLOGICAL AND EPIGENETIC FUNCTIONS FROM A MASS SPEC PERSPECTIVE IT'S ACTUALLY AGAIN QUITE A ONNARY PTN, MAKING THEM DIFFICULT TO DETECT, AS DESCRIBED PREVIOUSLY AND THERE'S NUMBER OF INTERFERENCES THAT MAKE DETECTING THIS DIFFICULT AND THE MASS SHIP UPON SY TRAILIZATION IS .98 DALTON. SO TO GO AFTER THE PTM WE CREATED USING SOLID FACE SYNTHESIS APPROACH, THIS CITRULLINE DEPENDENT PROBE OF PROTEOLYTIC ACTIVITY SO SIDE CHAINS ARE NOW CITRULLINES. AND THERE'S A LITTLE PEPTIDE ON THE END BECAUSE TRIP SIP FAMILY PROTEASES ARE LIKE A CASH NEAL CARBON ADJACENT AND CREATED A LOOP # AND 2 GENE LIBRARY OF TRYPSIN, AND WE COMPARTMENTALIZE THESE IN VITRO TRANSCRIPTION LANCELATION AND SEPARATED BEADS OUT BY FLUORESCENCE ACTIVATED CELL SORTING. THESE ARE BEADS BEFORE GOING INTO EMULSION IBTT AND AFTER EMULSION THERE'S A POPULATION OF HIGHLY FLUORESCENT BEADS GENERATED, WE COLLECTED 53 OF THE MOST FLUORESCENT BEADS DEPOSITING AS SINGLE BEADS IN SINGLE WELLS OF PCR TO AMPLIFY THE DNA OFF THE BEADS. FROM THERE WE HAD TO GENERATE AN ASSAY THAT WILL TELL US WHETHER OR NOT THE PROTEOLYTIC ACTIVITY WE ARE INTERESTED IN IN ANY ONE OF THESE WELLS. SO I'M SHOWING YOU THE MOST PROMISING MUTANT THAT WE HARVESTED FROM THESE ASSAYS TO DESCRIBE HOW THE SYSTEM WORKS. SO WE HAVE A WHOLE SERIES OF DIFFERENTLY ACTIVE CITRULLINE MUTANTS TO EXPLORE. MUTANTS ARE SHOWN IN ORANGE AND WILD TYPE TRYPSIN IS SHOWN IN SIGH YEN. THE ASSAY WORKS AS FOLLOWS WE PERFORM IT IN PLACE AND QUANTITATIVE PCR INSTRUMENT BUT WE BASICALLY PLACE CITRULLINE PROTEOACTIVITY ALL FOR IN VITRO TRANSLATION AND DNA FROM WILD TYPE OR ONE MUTANT IN THE WELL AND WE SHOVE PLATE IN THE QUANTITATIVE AND WATCH TRANSCRIPTION AND TRANSLATION HAPPEN IN REAL TIME. SO WHAT THAT LOOKS LIKE IS FOR THE NEGATIVE WE GET A BASELINE. SO NO DNA ADDED TO THE REACTION. FOR TRYPSIN BEING TRANSCRIBED IN TRANSLATED IN THE PRESENCE OF CITRULLINE WE GOT NO ACTIVITY BECAUSE THERE IS THE TRYPSIN HAS NO MEASURABLE ACTIVITY AGAINST CITRULLINE AS SIDE CHAIN BUT THE MOST PROMISING EXHIBITED HIGH ACTIVITY AGAINST CITRULLINE PROBE THAT CAUSED US TO HARVEST, OUTSEQUENCE AND EXPRESS IT. WE CAN DO THIS ANALYSIS USING THE TRIP TICK ANALOG OF THAT PROBE AND UNSURPRISINGLY TRYPSIN EXHIBITS HIGHLY ROBUST PROTEOLYTIC ACTIVITY AGAINST THE TRIP TICK PROBE, ALSO NOT SURPRISING MUTANT TYPES AN RESIDUAL BACKGROUND TRIP TICK ACTIVITY IS AFTERALL DESEN DEBIT OF TRYPSIN AND DURING EVOLUTION EXPERIMENT WE DID NOT COUNTER SCREEN AGAINST TRIP TICK ACTIVITY SO CAME ALONG FOR THE RIDE. THERE WAS AT THIS POINT IN TIME WE WERE REALLY EXCITED AND DECIDED TO RECOMBINANTLY EXPRESS PROTEASE AND LOOK IN A MASS SPEC EXPERIMENT BECAUSE THAT WAS THE AIM OF OUR INNOVATOR GRANT. THIS WAS AN EXCITING DAY. SO WE TOOK OUR PROTOIATION AND WE LOOKED AT THE ENZYME PROTEIN ARGININE DEAMINASE BECAUSE LIKE KINASE, IT NOT ONLY CATALYZES THE SY TRILL NATION OF OTHER PROTEINS BUT AUTOSY TRILL LYNNATES. SO WE DID A QUICK MASS SPEC EXPERIMENT DIGESTING THE AUTODEAMNATED FORM OF PAD 4 WITH MUTANT PROTEASE WHICH CLEAVES PROTEIN IN CITRULLINE AND OTHERS OR WITH TRYPSIN YOU CAN SEE MORE FEATURES IN THE MUTANT MASS SPECTRUM THAN THE TRIP TICK SPECTRUM. WENT UP ONE LEVEL REGION WHICH WE GOT EXCITED, SINCE I'M SPECTROSCOPIST SESSION HERE EVERYONE IDENTIFY WITH ME WHEN WE'RE EXCITED ABOUT PEAKS. SO THIS IS SPECTRUM FROM THE TRYPSIN DIGEST AUTODEAMNATED PAD 4 AND THIS REGION OF THE SPECTRUM, THERE ARE FEW FEATURE, THOUGH WE DIGESTED THE NATIVE FORM OF PAD 4, PRIOR TO DEAM NATION, WE SAW THESE TWO NEW FEATURES AND WE GOT REALLY EXCITED WHEN WE SAW THE MUTANT DIGEST OF THE AUTODEAMNATED PAD 4. NOT BECAUSE OF THE FEATURE OVER HERE AT 1387, THAT IS SEEN IN BOTH THE NATIVE PAD 4 AND THE AUTODEAMNATED PAD 4 INDEPENDENT OF SY TRILL NATION. OVER HERE WE SEE A FEATURE THAT UPON DEAM NATION EXHIBITS A ONE -- ESSENTIALLY ONE DALTON SHIFT IN THE MAIN ISOTOPIC IN THE MONOISOTOPPIC MASS. SO WE THOUGHT MAYBE THAT MIGHT BE OUR FIRST BAY BURKE FIRST CITRULLINE DEPENDENT DIGEST PRODUCT. WE CAN DO A THEORETICAL DIGEST OF THIS PROTEIN AND SPECULATE THAT THIS 1400 M OVER Z PEPTIDE WAS THIS SEQUENCE. WHICH WOULD MAKE SENSE IF THE 14 L 1 PEAK WAS A CITRULLINE PRODUCT IN THE ORANGE IN THE ACTIVE ENZYME KNOWN IN THE LITERATURE TO BE SY TRILLNATED. SO THAT GIVES US CONFIDENCE WE CAN DROP A SMALL BUCKET OF MONEY ON A HIGH RESOLUTION LCMS AND MS EXPERIMENT AND POTENTIALLY GO HUNTING FOR THESE TWO PEPTIDES IN HIGH RESOLUTION DATA. SO WE DID AN EXTRACTED ION CHROMATOGRAM ANALYSIS, 1400 OAR 1401 AS A FUNCTION OF LC ONE, ABLE TO OBSERVE TWO CHROMATOGRAPHLY DISTINCT SPECIES THAT CORRESPONDED TO THE TWO PEPTIDES, ONE AT 1400, IT'S OBSERVED AS TRIPLE A CHARGED SPECIES, AND THE OTHER AT 1401. THE MASS DIFFERENCE CORRESPONDED EXACTLY TO CITRULLINEATION. AND ACTUALLY MUCH TO MY EXCITEMENT, WAS THE DATA AND THE MS 2 SO WHEN WE LOOKED AT THAT CITRULLINEATED PEPTIDE THE SECOND DILUTED PEAK HERE IN THE MS 2 SPECTRUM, WE ACTUALLY FOUND THAT THE Y 1 ION WHICH IS CLEAVAGE OF THE TERMINAL MI BOND WAS PRESENT. THAT CLEAVAGE EVENT IS IMPORTANT BECAUSE THAT ION CORRESPONDS TO CITRULLINE MASS, 176.103. THAT ALLOWED US TO UNDERGO AND DIVE THROUGH ALL THE MASS SPEC DATA FROM THIS RUN WHICH IF YOU'RE NOT FAMILIAR WITH IT AS GIGABYTES OF DATA. AND FISH OUT OF A SMALL HANDFUL OF SPECTRA WITH 176.103 SIGNATURE OF CITRULLINEATION. THESE ARE THE SPECTRA AND WE PAIRED THEM WITH THE EXCEPTION OF ONE, ALL OF THEIR UNMODIFIED FORMS AND DO TWO PEAK EXTRACTED ION CHROMATOGRAM ANALYSIS AN UNAMBIGUOUSLY IDENTIFY YOU WILL SITES OF CITRULLINEATION IN PROTEASE. IN THESE PARTICULAR PEPTIDES AND MAPPING TO ARGININE RESIDUES ON PAD 4. SO THIS PROTEASE WITH ITS NEW BACK ALLOWED US TO SEARCH FOR THE NEW SEED AND GIVES US HOPE THEN THAT WE CAN GO TO APPLY THIS TO OTHER PROTEINS WHICH WE DID, WE ACTUALLY ENDED UP APPLYING THIS TO FIBRINOGEN, A TARGET OF PAD 4 CITRULLINEATION AND IDENTIFIED 25 SITES OF CITRULLINEATION ON IT ON TWO NOT KNOWN IN THE LITERATURE PREVIOUSLY. WE NOW HAVE LEADS ON SEVERAL OTHER POST TRANSLATIONAL MODIFICATION DEPENDENT PROTEASE, SOUPED UP OUR EVOLUTION PROCEDURE TO HOPEFULLY DISCOVER OTHER PROTEASES INCLUDING PROTEASES THAT WOULD GIVE US THAT MAGICAL OTHER FRAGMENTATION PATTERN FROM TRYPSIN TO GIVE FULL COVERAGE AND NOT PTM DISCOVERY MISSIONS BUT FULL PROTEIN SEQUENCING MISSIONS. SEPARATELY I WOULD LIKE TO THANK THE INNOVATOR AWARD IN PARTICULAR, FOR OUR ABILITY TO NOT ONLY EXPLORE WORK THAT I PRESENTED TO YOU TODAY, BUT OUR LABORATORY WAS ALSO ABLE TO INITIATE A COMPLETELY DIFFERENT LINE OF INVESTIGATIONS THAT INVOLVED DNA AND CODING OF THINGS ON BEADS, WE STARTED A PROGRAM TO LOOK AT DNA ENCODED SOLID PHASE COMBINATORIAL LIBRARY SYNTHESIS AND ULTRA MINIATURIZATION OF ASSAYS IN MICROFLUIDIC DROPLETS TO RECONSTITUTE ALL OPERATIONS OF HIGH THROUGH PUT MOLECULAR SCREENING CENTER WHICH WE HAVE ONE AT SCRIPPS IN FLORIDA ON TO A TINY DEVICE. THAT WORK WOULDN'T HAVE BEEN POSSIBLE WITHOUT THE FUNDING FROM THE NEW INNOVATOR PROGRAM. SO WITH THAT I'LL CONCLUDE BY THANKING THE PEOPLE IN MY LAB, VALUE CABOT, MY RESEARCH TECHNICIAN WITH ME FROM THE BEGINNING IN THE TRENCHES OF TRYING TO FIGURE HOW TO TRANSCRIBE AND TRANSLATE TRYPSIN IN VITRO. AND LATER ON WE WERE JOINED BY INCOMPARABLE DUCK TRAN AS POST-DOCTORAL FELLOW, SHE HAS HER OWN LAB IN VIETNAM. ALSO I WOULD LIKE THE THANK IN ADDITION TO NIH FOR FUNDING MY LABORATORY IS ALSO VERY FORTUNATE THROUGH FUNDING FROM NSF AND DOOR PA FOR EFFORTS IN ALL THESE AREAS. THANK YOU VERY MUCH FOR YOUR ATTENTION AND FOR THIS SYMPOSIUM, I'LL TAKE QUESTIONS. [APPLAUSE] >> I CAN SEE HOW METHODS ARE FOR DESIRED CLEAVAGE SITES BUT IT DOESN'T NECESSARILY SELECT A GAZE AN DESIRE AT ONCE. DID YOU FIND ADDITIONAL ACTIVITY OF YOUR PROTEASES FOR CLEAVING AFTER OTHER AMINO ACIDS WHICH IS UNDESIRABLE? >> WITH THIS PROTEASE IN PARTICULAR? >> YES. SO WE APPLIED SOME WELL KNOWN LITERATURE AFTER WE HAD THE DATA SET FROM FIBRINOGEN WHICH ALLOWED US TO HAVE MORE CLEAVAGE JUNCTIONS TO ANALYZE, WE WENT THROUGH AND DID CLEAVAGE JUNCTION ANALYSIS OF PROTEASE. I TURNS OUT HERE IS CITRULLINE HERE AND WE LOOK AT CLEAVAGE JUNCTION OBSERVED IN THE CITRULLINE PROTEASE AND THEN THE WILD TYPE ITSELF. THE MAJORITY OF THE CLEAVAGES ARE STILL ARGININE AND LYSINE BUT OUR PROTEASE HAS INTERESTINGLY PICKED UP DEPENDENT CLEAVAGE ACTIVITY AND TYROSINE DEPENDENT CLEAVAGE ACTIVITY. THOSE WERE NOT PART OF OUR ORIGINAL SCREEN BUT WHEN WE DID THIS EXPERIMENT, THESE DATA WERE QUITE VALUABLE BECAUSE THEY ALLOWED US BACK IN THE DATABASE SEARCHING AND CHANGE ENZYME DEFINITION WHICH ALLOWED US MORE PEP SIDES. IT TURNED OUT THIS WAS TOTALLY SERENDIPITOUS BUT TURNS OUT THAT THOSE EXTRA CLEAVAGES ALSO ENABLE US TO JUMP OUR SEQUENCE COVERAGE BY ALMOST 20% IN SOME CASES FOR SOME OF THE PROTEINS THAT WE WERE LOOKING AT SO WE INADVERTENTLY GOT FULL SEQUENCE COVERAGE ISSUE BY OPENING UP THESE NEW CLEAVAGE ACTIVITIES. >> SO YOU DON'T SEE INCREASED SEARCH SPACE AS BEING DETRIMENTAL FACTOR? >> IT IS WORKING WITH A MORE COMPLEX SAMPLE SO THE EARLIER SAMPLES WERE PURIFIED PROTEINS. WE AREK LOUING NOW WE HAVE A DATA SET OF MOUSE BRAIN HISTOLOGICAL SECTION WHICH MIGHT HAVE BEEN A FEW JUMPS TOO FAR FROM A PURIFIED PROTEIN. BUT WE LOOK AT THIS TO SEE IF WE CAN MINE THROUGH WITH THAT CITRULLINE SEED THAT I TOLD YOU ABOUT EARLIER. >> COOL. THANK YOU. >> HOW MUCH DOES PROTEIN STRUCTURE LIMIT THE ABILITY OF THESE PROTEASES TO SAMPLE SITES AND HAVE YOU THOUGHT ABOUT DOING JUST LIKE A TRYPSIN DIGEST? AND THEN A SECONDARY DIGEST FOR THE PTM, DO YOU THINK YOU WOULD PICK UP SITES YOU WOULD MISS BECAUSE THEY'RE PRECLUDED BY PROTEIN STRUCTURE? >> ANSWER YOUR FIRST QUESTION, DOES THE PROTEIN STRUCTURE INFLUENCE CLEAVAGE SITES? WE DO DIGESTIONS, WE DIGEST PROTEINS THAT HAVE BEEN ALKYLLATED REDUCEND AND UNFOLDEDDED SO THE STRUCTURE IS OUT OF THE QUESTION. YOUR SECOND QUERY ABOUT WHETHER OR NOT WE OPEN UP MORE COVERAGE BY DIGESTING WITH OTHER PROTEINS, WE HAVEN'T TRIED IT YET BECAUSE THIS PROTEASE BRINGS WITH IT TRIP TICK ACTIVITY FOR FREE. BUT I WOULD TOTALLY TRY THAT IF WE HAD A PTM DEPENDENT PROTEASE THAT ONLY CLEAVED AT PTMs. WHICH WE WOULD LIKE TO GET. >> SO THAT CONCLUDES THIS AFTERNOON'S SYMPOSIUM. I WOULD LIKE TO THANK ALL OUR SPEAKERS AGAIN. [APPLAUSE] >> THERE IS A -- THE POSTER SYMPOSIUM IN THE ATRIUM OUT HERE AND WE INVITE EVERYBODY TO SPEND TIME LOOKING THROUGH THOSE POSTERS AND ALL OF THIS AFTERNOON'S SPEAKERS SHOULD BE AROUND AS WELL. SO I -- >> GREAT, THANKS A LOT. LET'S THANK ALL THE SPEAKERS AGAIN. AND WE WILL BEGIN TOMORROW 8:30 HERE. THANKS.