Alzheimer's: commentary on treatment strategies

My goal is that any reader can take something away from this Alzheimer's post, regardless of their experience in medicine or science. It's Friday Links-driven, but with far more of my own commentary/explanation than normal.

First up: an article about a clinical trial for Alzheimer's treatment that appears to have failed. While I don't research Alzheimer's, I'm going to argue in this blog post that 1) single "magic bullet" treatments are unlikely to ever work for a chronic complex disease that develops over many decades like Alzheimer's (once the disease has actually manifested symptomatically), but 2) single "magic bullet" prevention methods might work for specific patient sub-populations. In this case, different prevention methods would work for different sub-populations. Meanwhile 3) Broad, non-specific treatments that target multiple biological processes are better for stabilizing Alzheimer's after it's been diagnosed (discussed in the second link). Note that this is NOT the same as combination therapy.

http://www.technologyreview.com/view/428592/pfizer-disappointed-with-first-results-for/?ref=rss

Bapineuzamab is an antibody that recognizes and binds to beta-amyloid, one of the molecules involved in the pathogenesis of Alzheimer's (note that I don't say it's the cause or even a significant cause). For my non-biomedical readers, this is a common treatment strategy nowadays. Say protein X causing disease Y is floating around in the space between cells in your body. Specific antibodies can be made to bind specifically to protein X which physically blocks protein X from damaging other things ("neutralization"). Furthermore immune cells are then able to recognize the antibody bound to protein X, and clear protein X from the body. This strategy is used in treatment of diseases like rheumatoid arthritis and lymphoma, and it works wonders.

But Bapineuzamab failed to have any effect on Alzheimer's progression in this Pfizer trial. Why? I'm going to use my extraordinary powers of hindsight (dig deep- you probably have this ability too!) and say that it probably has far more to do with the fact that beta-amyloid is just a tiny piece of the puzzle for Alzheimer's than any problem with the drug. Bapineuzamab probably recognizes beta-amyloid just fine, and it might even clear beta-amyloid from the body. But I doubt that clearance of beta-amyloid would have any effect on Alzheimer's. Why? Because it's too late.

The trial looked at treatment of early-to-moderate Alzheimer's, but Alzheimer's develops over decades. For a while, it's just Mild Cognitive Impairment (MCI), but lots of people get MCI and never progress to Alzheimer's. So figuring out a way to predict Alzheimer's progression through things like blood tests and brain imaging is all the rage now (see third link). And that's why the trial focused on Alzheimer's rather than pre-Alzheimer's. But in the patients that do progress, what's going on?

The length of time that it takes for Alzheimer's to develop means that, even if beta-amyloid were to be the ultimate cause, then beta-amyloid can trigger numerous other biological processes that are themselves damaging to the brain. By the time someone has Alzheimer's symptoms, these "secondary processes" are already robust so more damage is occurring independent of beta-amyloid, and a lot of neurons are already malfunctioning or dead. This fundamentally alters the biology of the brain so that treatments are unlikely to reverse anything, and multiple causes of degeneration make it unlikely that single treatment would slow the progression of the disease. I won't review the ginormous body of literature implicating all sorts of things in Alzheimer's pathogenesis, but I'll take a couple as an example.

One way to abstract Alzheimer's (see picture below) is that everything causes everything else. It's a complicated feedback loop (or feedback web) where a bunch of biological processes all cause and worsen each other over a period of many years. For example (highlighting a tiny portion of the feedback web), extracellular amyloid or intracellular tau (hallmarks of protein misfolding) acting on one subset of neurons might interfere with normal breakdown of neurotransmitters, as well as directly causing neurons to fire inappropriately. Too much excitation of nearby neurons results in excitotoxicity (killing those cells) or in inappropriate activation. The brain might remodel in reaction, forming new synapses that result in electrical feedback loops that reinforce each other. The resultant clinical and subclinical seizures might interfere with brain function long after the damage is done. As neurotransmitters start to diffuse inappropriately due to synaptic dysfunction, they start affecting other cells indiscriminately, and damage may occur to brain's extensive blood system. This allows in immune cells that further alter the blood vessels to essentially break down the blood-brain barrier.  This lets in various molecules that again might cause excitotoxicity or protein misfolding, and maybe it even lets in bacteria. So damage leads to biological response that causes more damage, leading to further responses etc.

All of these processes cause each other, and all of them cause disease. Targeting a single initiating factor (which varies from patient to patient) might work for prevention, but not for treatment.

Importantly, many of the damaging processes are variations of normal brain biology. For example, microglia (kind of like the brain's immune system) see damaged neurons and eat them up. If you remove the source of damage, it is perfectly possible they continue eating up neurons instead of letting the neurons recover after the damage. There are medical examples where a disease has manifested for so long that cells that normally act to ameliorate a disease are "locked in" to their action even when it's no longer necessary, and they end up causing damage themselves (for example tertiary hyperparathyroidism). I think of this as part of the more general inflammatory response that occurs whenever there is damage- all sorts of immune cells react to initial damage and can end up causing more damage than the original insult (if you think that makes no evolutionary sense, it actually does. I might expand on that in a future post).

What about prevention? Note that things like excitotoxicity, brain remodeling, and breakdown of the blood-brain barrier can in turn cause beta-amyloid buildup. So there's no reason why beta-amyloid had to be the initial insult- in many (most?) patients beta-amyloid is probably secondary to another biological process. In these cases, drugs targeting beta-amyloid production and degradation probably would not have any preventative effect, because beta-amyloid was not the initial cause. However, there are subsets of patients where beta-amyloid is implicated as a major genetic cause (mutations that affect beta-amyloid production like in Down's syndrome, ApoE4, and presenilin). Prevention using Bapineuzamab is conceivable in those patients, as it would stop the secondary processes from occurring in the first place. However, if we found in another sub-population that inappropriate inflammation due to immune system malfunction (kind of like an autoimmune disease), then prevention would involve anti-inflammatory drugs (Aspirin? IVIg? Prednisone?). Thus, preventative measures for Alzheimer's would be specific to the patient's initial cause(s) of degeneration, which can vary widely depending on the patient.

http://www.technologyreview.com/news/428546/study-suggests-alzheimers-disease-can-be/?ref=rss

On the bright side, a very small trial showed stabilization of Alzheimer's with treatment with IVIg (intravenous immunoglobulin). IVIg is simply the mix of the collection of antibodies isolated from multiple human volunteers. There are antibodies against everything- bacteria, toxins, some human proteins, etc. The authors here speculate that there is an antibody targeting amyloid-beta, tau, or some other molecule. While this might be part of the picture, I worry that researchers might go after specific antibodies (which is just like the above Bapineuzamab trial). I challenge the notion that IVIg's broad and non-specific effects are a disadvantage. While one might think it's not 'optimized' for Alzheimer's treatment, the very fact that Alzheimer's involve a complicated web of numerous biological processes means that we need to target them all at the same time. Thus, a non-specific treatment with numerous antibodies doing many different things might in fact be the key to IVIg's efficacy.

For example, IVIg is used in the treatment of autoimmune disease, dampening down immune responses. How it accomplishes that is unclear (and is a bit counter-intuitivee since antibodies MEDIATE the immune response), but it is believed that it both interferes with the specific endogenous antibody that causes disease, as well as flooding the system with so many antibodies that it diverts the immune system from inflammation. Perhaps IVIg is dealing with the inflammatory component of Alzheimer's? Perhaps it interferes with the endogenous cells/antibodies that are damaging the brain? Thus, trying to narrow down the treatment to a single antibody or a few antibodies would eliminate some of the broad effects of IVIg that would be critical for influencing the numerous biological processes. This is different from combination therapy because we're looking for one or two treatments to influence many things (100+) things at once, rather than multiple (3-5) treatments for multiple (3-5) things.

Also note that IVIg only stabilized the disease, it didn't reverse anything. That's because the damage is done- the neurons are dead and the brain has remodeled itself. At this point, Alzheimer's could only be reversed by making new neurons by stem cell therapy. Because many of those dead neuronal circuits likely encoded specific memories and personality traits, we would need to find a way to program those back into the new neurons. Those would be Sci-Fi technologies that haven't even been imagined yet.

http://www.technologyreview.com/news/428480/an-alzheimers-warning-25-years-before-symptoms/?ref=rss

More on prevention: you need to be able to predict who is going to get the disease in order for a prevention to fulfill a cost-benefit analysis (since preventative treatments might have their own side effects, and you don't want to expose people who will never get the disease to unnecessary risk). This brain imaging study suggests that this is possible, at least in one inherited subtype of the disease.

http://www.newscientist.com/article/dn22128-alzheimers-villain-cures-multiple-sclerosis-in-mice.html

I'll just leave with you an interesting tidbit- they used beta-amyloid injected into the body cavity of mice to reverse multiple sclerosis (MS). What? Isn't beta-amyloid bad? But this sort of goes with my idea that injecting IVIg "distracts" the immune system from attacking brain cells, just like beta-amyloid might "distract" the immune system from attacking myelin sheaths in MS. I think the lesson here is: We need to think outside of the box and consider counter-intuitive treatments to deal with these complex diseases.

Friday Links: Online education, crime algorithm, health care

I was looking at my Blogger stats and discovered that someone found my site by googling "help i'm addicted to wasting time on the internet." I googled it and asked an acquaintance to do the same (since Google results are personalized), and it turns out my article on Internet time-wasting is hit #3-#5, depending on who searches for it. Nice. Interestingly, it goes down with "i'm addicted to wasting time to the Internet," and doesn't come up if you just google "addicted to wasting time on the internet."

http://www.nytimes.com/2012/07/20/opinion/the-trouble-with-online-education.html?_r=1&smid=fb-share

Those who started reading my blog from the beginning know that I started off by explaining an interesting example from an online course on Coursera. To me, online courses are one way of exposing my mind to new ideas that break my preconceptions and defy conventional wisdom. Letting go of long-held ideas is critical to getting anywhere in science. But while I currently treat online education as a fun side project, there is a good debate going on as to what the role of online education will be in general for K-12 and college. Since it's the new thing, everyone wants to know to what extent it will replace traditional classroom learning. In this NY Times article Mark Edmundson, a professor at University of Virginia appears to argue against the widespread use of online education, saying there's nothing you can get from an online course that you can't get from a good book. I largely disagree.

I agree only with his thesis, "But can online education ever be education of the very best sort?" Well sure, it can't be the BEST by itself, but it sure beats the average educational experience in the US. In my opinion, education isn't about learning facts, it's about learning how to make arguments and how to understand and solve problems. And yes (agreeing with Edmundson), this can't be accomplished in a one-way didactic lecture. It needs dialogue and requires students to get to the answer themselves, with proper cultivation from the professor. However, I disagree that this says anything about the value of online education in the grand scheme of things. I don't think online education is meant to completely replace classroom learning, and certainly won't replace the best professors at the best universities. Learning how to solve problems requires some starting facts (and in science, LOTS of starting facts), and those facts should be communicated in the most efficient and organized manner. When done properly, one-way didactic lectures synergize with books (rather than being redundant). And no, I don't find it likely that every teacher individually optimizes the communication of those facts. Furthermore, in many school systems this is so inefficient that they spend all their time lecturing facts and no time on critical thinking. Only a few lucky students get real dialogue learning, since you inherently need small classroom sizes for that. Thus online education, taught by THE best educators (like, the best in the world), would go a long way to improving K-12 education. Then teachers can focus on problem solving sessions rather than lecturing facts.

I envision two parts to future education (both K-12 and college): 1) one-way lectures taught by the very best people who have developed the best ways to explain something. These can be online and available for everyone in the world. 2) actual teachers or TAs that focus entirely on face-to-face dialogue. They don't provide any actual information- they present a problem and work with the students to reason through the problem, using information that they learned in the lectures. They are more like older colleagues than anything else. In college, I often learned way more in small discussion sections than in lecture. Lectures are necessary but not sufficient for education.

In fact, when I served as a TA, there was one module that I didn't know anything about. So what did I do? I studied it just enough to get an intuitive feel for it, and then I just pummeled my students with questions while working through problems. I didn't provide a single answer for them (because I didn't know how to solve the problem), and if they asked a question I just asked a question back. The result? In my student evaluations, they specifically mentioned how well I taught that module. Real teachers don't need to know the answer.

http://www.technologyreview.com/news/428354/la-cops-embrace-crime-predicting-algorithm/

A crime-prediction algorithm takes crime data and balances information on the day, week, month, year, and decade scale to figure out where crime is statistically most likely to strike next in the city. Maybe a certain part of the city sees more crime frequently around the holidays, for example. Humans can't physically process and balance all of the data, so let a computer do it. This leaves more time for humans to do what they do best- interact with other humans. They show up, talk to people, and just by having a presence decrease crime.

http://ayoungmomsmusings.blogspot.com/2012/07/how-i-lost-my-fear-of-universal-health.html

Blog of a die-hard conservative Republican who moves to Canada and of course fears Universal Health Care. However, she soon discovers it's great and that more government control = more freedom for individuals to choose. A far more complicated issue than I'd want to address in a Friday Links entry.

Other random links:

http://www.newscientist.com/article/dn22088-heart-muscle-helps-cyborg-jellyfish-come-alive.html?DCMP=OTC-rss&nsref=online-news

http://www.huffingtonpost.com/2012/07/07/pacu-testicle-eating-fish-illinois-lake_n_1656015.html?utm_hp_ref=mostpopular

The future: no more secrets

The Internet lets people share the problems that are befuddling them. At the same time, the Internet allows for the dissemination of information for other people to solve those problems. And there's nothing stopping those people from connecting, other than the potential mistrust. I think one tantalizing idea is that we're moving towards a world where everyone (including companies) will share their information freely on the Internet. In other words, maybe everything will become open access. Nothing proprietary.

In the TED talk below, Don Tapscott tells the story of a gold prospector has collected a bunch of data on a geological site that he is evaluating. However, the geologists that he works with aren't able to locate the gold and aren't able to make recommendations as to where to start digging. So he thinks: maybe someone else would be able to figure it out. So he does something that is unheard of in business: he decides to publish his data online for everyone to see and held a competition for someone to locate the gold. The result? Someone found the gold, told him where to dig, and he made a bazillion dollars. Could he potentially have gotten scooped? Maybe someone would have sat on his result until the gold prospector gave up and sold his rights to the land, and then swooped in to grab the gold. But I think that's highly unlikely, since each person is competing against EVERYONE else on the Internet. If the malicious person decided to wait, then some other person would probably figure it out and then win his share of the gold in a FAIR manner. In this hypothetical Internet world where people freely share their "trade secrets," people who try to take advantage of other people will not thrive.

I think this is applicable to any field, especially science. If a PI needs something that doesn't exist yet but will likely require expertise outside their field, they currently have two options: hire a guy to work in the lab and develop it, or try to convince another specific PI to collaborate with them. The viability of both options is influenced by factors such as who is in your social network (i.e. having the right connections), investing time to find a person you can trust, and investing time to convince that person that it's a worthwhile pursuit. It's a lot of work and it is highly unlikely that you've found the ideal person for it, especially since you are not the expert and you aren't yet sure what is required to solve the problem. The biggest problem is that the PI is unlikely to disclose any detailed information about the project until he has found the person he wants to work on it. But this is totally backwards. You don't know if a person is going to be able to solve a problem until they've actually looked at the problem. Thus, if someone needs outside help, they can simply release all the relevant information onto the Internet and look for the guy who can solve it. Getting scooped is not an issue because everyone on the Internet already knows exactly what you did and what you contributed. (By the way, I think peer-review is going to get crowdsourced in the future, so it won't matter anymore who is first to publish a result in some paywalled journal). In this system, every professional scientist (or non-scientist, for that matter) could do a little science freelancing on the side, and personally I think it would be fun as hell. 

This also applies to companies. Right now, there's little doubt that the current pharmaceutical industry model sucks. What follows is a simplification, but one issue is that each company sits on a wealth of proprietary information but usually has insufficient power to utilize it. Note that these data sets are ASTRONOMICAL- drug libraries, clinical trial data, synthesis methods, preclinical data that shows that X drugs affect Y biological processes, etc. etc. There's no way that the employees of one company are going to use those data sets to their full potential. The companies are waiting for the chance to make money off of it, but because they have insufficient brainpower to tackle massive data sets, many drugs are not directed to their "ideal" patient populations, so many of the drugs fail. Plus, companies use the patent system to actively prevent others using that information, even if they independently discover it. And because the patent system is not perfect, everyone wastes time suing everyone else. Why bother with all the secrets? I think a lot more drugs would be successfully developed if every person in the world could look at pharmaceutical data and make their suggestions as to which drugs are promising for what diseases.

Remember, the goal shouldn't be to beat other people to the right answer. The goal is to find the right answer. Secrets were viable in the past because problems were simpler. Science didn't involve massive amounts of data. A small group of people could solve the problems without letting anyone else know what they're doing. But no more.

Please comment on my naivety.

The scale of neurology is larger than the observable universe

This week, I am happy to report that I am pretty much back on track in terms of my work routine, and productivity is once again through the roof. Spelling out and publicizing my exact strategy for not wasting time on the computer definitely motivated me to adhere to it- which is the entire point of writing a blog! But I also have another motivating factor working in my favor this week- a new undergraduate just joined the lab! When I teach, I feel like I learn more than when I'm just trying to learn. Like writing, teaching someone forces me to spell out my logic as clearly as possible, which both clarifies my thinking and uncovers hidden assumptions I had been making that may be wrong. Furthermore, the best way to teach is to ask the student questions that stimulate the student's thought and lets the student figure out the answer. That means I need to put myself in the position of a beginner, so I learn more about my own field and gain insights that I might have missed. It's true that until you can teach a subject, you haven't mastered it. But what is often overlooked is that the very act of teaching something is the METHOD by which one masters something.

Yesterday, I attended the Life Sciences Institute symposium, and this year's focus was on Neuroscience. The speakers were amazing (most were HHMI), and it reminded me why I want to go into neurology. Let's just examine some of the highly attractive intellectual aspects of neurosciences

Brain-specific systems: Molecular and cellular mechanisms that are unique to the brain and, in many cases, unique to the human brain. Non-coding DNA seems to be one of the biggest things that separates us from chimpanzees- why? A lot of it might be transposons that are specifically activated in the brain to jump around and disrupt genes, so that every brain cell has different DNA- thus creating a diversity not seen in any other organ system, other than the immune system. Another thing, which I just learned from Robert Darnell, MD, PhD (Rockefeller) is that the brain has its own splicing system (Nova proteins, etc) allowing the generation of novel isoforms not seen anywhere else in the body. 20,000 genes becomes 100 or 1000 times that number because the brain can generate far more unique proteins than the rest of the body due to novel mechanisms of splicing. Furthermore, different parts of the brain have different splicing systems, and in fact different parts of the same brain cell have different splicing machinery- possibly explaining some aspects of memory assuming these are stable states. And so it also makes sense that certain cancers would co-opt the Nova system to drastically change their gene expression profiles and give them a proliferative advantage, despite the immunological risk it puts the cancer at (spontaneous regression of Nova+ cancers have been observed due to the immune response).

Combinatorial complexity: This point can be best illustrated using one of the simplest examples in neuroscience (even though really it's not simple at all). The problem is recognizing self vs. non-self. Neurons don't want to synapse onto themselves because otherwise they inhibit themselves and become useless, or they hyper-activate themselves and end up killing themselves. But how does a highly branching neuron figure out that the neuron it has reached is another neuron or another part of itself? Larry Zipursky, PhD (UCLA) has discovered how this is accomplished in the Drosophila fruit fly. The Dscam class of molecules have alternative exons at four positions. Combinatorial complexity means that 12 x 48 x 33 x 2 = 38,000 unique Dscam molecules. Furthermore, each neuron expresses a random combination of about 50 different Dscam molecules. How many different profiles thus are possible in the brain? 

38,000^50 = 10^229. That is far greater than the number of particles in the entire universe. In fact, if every particle in the universe had an entire universe inside of it, and every particle in that universe had a entire universe inside of it, 10^229 is still far larger. Now, when a Dscam group on one neuron encounters a Dscam group on either the same or another neuron, it only binds if they match sufficiently. If it binds, they inhibit each other and cause the synapse to fail. So essentially, the likelihood that two different neurons will have profiles similar enough to inhibit each other is essentially non-existent. Thus, every neuron has a unique barcode that allows its dendrites to recognize other dendrites on itself. Even cooler, the Zipursky lab systematically deleted alternative exons until they figured out how many unique Dscam molecules are required to prevent inappropriate self-synapsing and inappropriate avoidance of non-self.

Region-specific features: Let's not forget that the brain is huge. Really huge. The human brain should not be thought of as one organ system. A single brain's complexity is more on the order of the entire rest of Earth's biosphere. So one part of the brain might act under totally differently principles than its neighboring part, even though the majority of proteins are the same. So when you treat the brain with a single simple drug, it may have really awesome effects in one part, but it's going to affect everything else too, possibly adversely. Let's take dopamine as an extremely simple example. Insufficient dopamine is a cause of some Parkinson's symptoms, so dopamine therapy can have massive benefit in terms of quality of life for Parkinson's patients. But dopamine is also inappropriately elevated in an entirely different part of the brain in schizophrenia, so a potential side effect is schizophrenic-like symptoms. Conversely, treating schizophrenic patients with dopamine antagonists can have Parkinsonian side effects. 

Another example: yesterday Luis Parada, PhD (MIT) discussed his work on SSRI anti-depressant therapy. He found that the reason why SSRIs take months to work even though they cause immediate serotonin changes is that SSRIs enhance hippocampal neurogenesis over time. New neurons need to form for SSRIs to work. More interestingly, exercise seems to have the same effect- explaining why I'm always happier after exercising consistently. Furthermore, activating hippocampal neurogenesis is sufficient to reverse depression and anxiety-like symptoms in mice, and blocking neurogenesis can block the positive effects of anti-depressants and exercise. This has major implications for depression therapy, since SSRIs currently affect serotonin all of the brain, resulting in all sorts of changes that may have all sorts of adverse effects. So if we develop a drug that specifically activates hippocampal neurogenesis, we can treat depression without the side effects. Exercise should also be incorporated as a mainstay of depression therapy. Lastly, I'd like to point out that this strategy can be used for cognitive enhancement in healthy people. Meanwhile, I'm going to keep on exercising.

One reason I want to go into neurology is that there are few good therapies for any of the major neurological disorders. But I have little doubt, based on what I've heard at research talks, that major neurological therapies will reach the clinic right around the time that I start residency. Right now there are some crazy flowcharts for figuring out which patients receive which therapies. But the brain is an entirely different animal. Figuring out which patients will benefit from neurological therapies (and cognitive enhancement) will require a fundamental understanding of these intellectually challenging topics such as combinatorial complexity. Thus neurology will provide me with intellectual challenge for my entire life. My prediction that most of the diseases of other organ systems will be cured within 100 years, and medicine will become tediously boring and trivial. I doubt that neurology will be solved for another 500.