A Deep Dive Into DNA Binding

How do transcription regulators bind DNA, and how does this binding in the test tube compare with binding in cells?

There has been a lot of speculation about special conditions inside cells that affect searching and binding by DNA-binding regulators. The fundamental sequence-specific affinity of a protein can be recapitulated in a test tube, (that is to say, "in vitro", even though test tubes are all plastic these days, not vitreous), but the cytoplasm / nucleoplasm is a different environment. The highly crowded nature of the environment is one issue. Another are specific factors like nucleosomes, chromatin structure, and the many other proteins that might compete with the binding protein on the DNA or alter the DNA shape, such as with twisting, bending, or chemical modifications. There are a lot of possible variables, and it has also been difficult to even assess quantitatively how well binding in the test tube compares with binding in cells.

A recent paper tries to delve into these issues in great detail, with rather confounding, if interesting results. I should say right off that this is a poorly written paper, with the kind of florid over-indulgence of figures that reminds me of my first poster. Anyhow, there were a few interesting points.

The regulator (or "transcription factor" as is historical / customary in the field) that they choose for this work is KLF1. This is a hugely influential regulator that orchestrates the maturation of erythrocytes from their progenitor cells, among other tasks. Knockouts of this gene leave mouse embryos dead at around 14 days, with profound lack of hemoglobin. Red blood cells undergo amazing development, ramping up hemoglobin production while they cut production of everything else, and finally eject their nuclei to become the streamlined and shapely pucks of oxygen-carrying goodness we are familiar with. KLF1 is part of a large family of proteins (that is to say, proteins that originated by accidental duplication from one long-ago parent through the history of life) that bind to DNA via three zinc fingers, shown as the gray bars in the diagram below. Zinc fingers are compact structures that bind to about three bases of DNA each, and, as the name suggests, contain a zinc ion that organize their structure. The human genome encodes over 8,000 zinc fingers. For a 3D look at a representative zinc finger structure in detail, go to the protein data bank.

Diagram of the protein domains found in SP/KLF family members, in humans. While these regulators all share a highly similar DNA-binding core of three zinc fingers, (gray bars at right), they are otherwise a riot of ancillary protein domains, which are generally involved in contacting other proteins of the transcription apparatus, forming cooperative complexes to either activate or repress transcription.

So KLF1 binds to a site of about nine informative bases, though these authors go on to show that a couple of extra bases on the sides are also significant to binding affinity. They do a thorough in vitro scan of all the sequence variations of KLF1's target sequence, coming up with a quantitative presentation of its "consensus", or ideal binding sequence, shown below. This sequence is rich in C's, and poor in T and G. 

KLF1 binds to this "consensus" sequence, via its three zinc fingers, as noted at the top. Each position of the binding sequence has its preferred and non-preferred bases, as indicated by the colored letters and their heights. 

They find that KLF1 has a pretty short residence time on its site, of about 2 minutes, thus is easily competed off by other DNA or other KLF1 proteins. Additionally, they look over the flanking DNA and find that not only do the two bases on each side of the consensus DNA sequence carry a lot of weight (presumably by directly binding KLF1 at its edges), but more distant flanking DNA has influential effects as well. These effects have a strong 10.5 base periodicity, indicating that they occur on one side of the DNA- the same side as the target binding site. But the preferred sequences at these distant positions are the opposite of the binding consensus, tending toward A and T. After considering a few hypotheses, the authors settle on a model where KLF1 searches for its binding site by hopping around on and near the DNA. The more semi-similar sites there are nearby, the more time this search takes, so it is better to bias the nearest locations against the consensus sequence.

But the main question they are asking is.. do all these in vitro parameters reflect also what is going on in vivo- in relevant erythroid human cells? For this, they use artificial DNA constructs with multiple KLF1 binding sites, and a methylation-protection technique where the bound protein physically "protects" its DNA site from methylation by an introduced enzyme (not to be confused with traditional DNA footprinting). These methylation sites are then read out by a complicated sequencing method, and displayed horizontally as shown below. Since the residence time of KLF1 is so short, the experiment was run as quickly as possible, with two minutes the minimum time point. Each of the KLF1 consensus sites (teal blocks, at top) were clearly protected from the methylase, while portions/lanes in between (15 bases long) were not, showing that these sites appear to have been specifically bound by KLF1 or one of its very close family members. This effect wore off quickly as KLF1 dropped off the DNA, (likely due to death of the cells/system they were working with, due to the methylation technique), and was hard to discern at fifteen minutes. So they conclude that the residence time of KLF1 is very similar between laboratory and in-cell settings. Incidentally, they show also that the consensus, flanking and distant flanking effects are also operative in cells, and thus the raw sequence-specificity of KLF1 binding to DNA is similar overall. 

"Footprinting" of KLF1 sites in cells. The DNA region is spread horizontally in these depictions, on the X axis, and white marks note methylation events, or non-protection from the methylase. The 15 bp borders between the KLF1 consensus DNA binding sites are white, while the sites themselves are dark or protected. In F, the control with mutated consensus KLF1 sites, there is protection all over the DNA, which the researchers interpret as a stable nucleosome plopped on this site instead of the specifically bound KLF1.

But suppose fewer KLF1 sites are used, instead of seven sites? What then? In vitro, binding of KLF1 is strictly linear with motif number (below). In vivo, on the other hand, binding is very poor to one or two sites, and only rises once three or more are available. The authors interpret this as showing competition against other binding proteins, principally nucleosomes (as above, right). And they also argue that this is not due to direct binding by the multiple KLF1 proteins vs the nucleosome, but is mediated by some other proteins (aka factors, in green, below), that bind to KLF1 and help to budge off local nucleosomes. KLF1 is a so-called "pioneer" factor which has this power, doubtless due not to its own structure alone, but due to the goons it calls in to help, such as nucleosome-displacing SWI/SNF or CHD1 complexes. 

In the test tube, affinity of KLF1 to its site is linear with the number of sites.

On the other hand, in cells, KLF1 has little occupancy until there are three or more sites available to bind (B, C). In D is shown the author's explanation, which suggests that large nucleosomes typically block the site and can only be evicted if several KLF1 binding sites and proteins cooperate to bring in some extra machinery (green).

Lastly, does all this work allow these researchers to predict where KLF1 binds within cells naturally? Not at all. They publish a fascinating analysis of the correlation between observed locations of KLF1 binding at its normal sites in normal cells (via the ChIP method) versus predicted sites in the genome from their in vitro analysis. And the correlation is pretty much zero (below). So KLF1 is only one actor, if an important one, in a flurry of activity going on in the nucleus, many aspects of which can supervene over its raw DNA binding preferences. Trying to predict all this, with an AI or other model, remains, as yet, quite out of reach.

Correlation between actual binding locations of KLF1 in normal cells (Y axis) vs predicted binding sites (X axis). Note the lack of correlation.

• Maybe things aren't so bad.

Binding Memories and Decision Context in the Temporal Lobe

Invasive detection of single human neurons allows analysis of their role in decision making.

Is AI as good as human brains? No and no- AI is not as energy efficient by a long shot, not as flexible, and particularly not equipped with sufficient world models (aka general common sense and context) to be dependable in decision making outside its narrow areas of expertise, such as coding and hacking through computer security barriers. But both are black boxes to some degree, with the brain being particularly hard to study, even for adepts.

A recent paper provided an example of advanced research in human brains. Sadly, it relies on people with epilepsy, whose cases are so severe that they are getting brain surgery to remove the epilepsy-activating locus. As they are already being given a highly invasive treatment, with electrodes probing likely areas of their opened brains in order to find the smallest possible excision site, there is a long history of researchers asking for access to these patients for unrelated studies in general neurobiology. This paper uses 16 such patients to read electrodes inserted into their medial temporal lobes, including the entorhinal cortex and hippocampus, areas well known to function in memory and its integration into navigation and other kinds of decisions. Fortunately or unfortunately, the entorhinal cortex is the most common activating nucleus location for temporal lobe epilepsy, the most common kind of focal epilepsy.

The testing procedure was relatively bare-bones. The subjects were first shown a large number of images to find responsive cells from among their implanted electrodes. Then the experimenters set up a task system. They would ask a question first, and the subject would have to remember that question as two pictures were then presented successively. These questions required some kind of comparison, such as "Which person shown is older", or "Which item is more expensive?". The idea was to track how the picture-responding cells (and other cells) activated while the items were being presented and the decisions were being made. But the researchers were particularly interested in other cells- those that responded to the question being asked, or perhaps responded to some combination of the question plus the image. In finding such cells, they could investigate how memory and subject binding between context and object is handled in the brain. 

The figure above shows the firing of individual cells. On the left is a cell that responds only to the image of a cookie. Other pictures (below) do not activate it, and nor do any of the context-leading questions, to any special degree. On the right is a cell that is quite special- it responds to a combination of one image (the train) and one question, but not any of the other questions, and not any of the other pictures. 

The figure above shows another kind of neuron firing- one that responds only to one question/context: is the person in picture one older than the one in picture one? But it does not respond to any picture in particular way, nor to the other questions. The dotty portions of the figure show the detailed firing of these cells, which is counted/binned/summarized in the graphs at the bottom.

This Venn diagram shows the counts of neurons with particular characteristics, from among the total of 3,109 neurons that were probed in these 16 patients. Due to the pre-experiment screening procedure, cells that respond to images (the "stimulus") were predominant. But cells that respond to the questions asked (the "contextual stimulus") also came up surprisingly frequently. In green are cells that respond to one or more questions only. Cells that responded to some combination of both picture and question are represented by the smallest circle, giving various possible computations of the decision being asked for. 

These graphs show a summary of the experiment through time, from where the question was asked, to the presentation of each picture, and to the decision being made about four seconds down the line. In the bottom, picture-specific neurons fire separately as they are stimulated by each picture in succession. What is interesting is the top graph, showing activation of context-stimulated neurons, not just as the question is being asked, but then re-activated as the subjects need to recall the question in the context of each picture. The control graph (red) is where the experimenters shuffled the data (pictures and questions) randomly, getting a net null level of activity. Note that there is a delay of about 40 milliseconds between the picture-driven activity peak and the question/context activity peak, indicating some time span for evaluation that is taking place, all followed by about a second until (on average) the decision finally arrives from the subject (at the end of the graph). In addition, note how the context neuron firing becomes (on average) even more intense as the trial progresses, from its original activation as the question is asked, through the presentation of each picture. This suggests that these neurons are not so much for storage of memories as for their activation and use in decision-making.

The main finding is that neurons corresponding to what is needed for a decision in this instance are clearly available if one looks carefully in the brain. There are neurons that "see" the picture, regardless of other distractions, neurons that "understand" the question, regardless of other pictures or questions (and remember that question later on), and there are neurons that get the binding of both conditions, between the picture and the question. Additionally, the researchers find that the stimulus neurons were primarily located in the entorhinal cortex, while the context neurons arose mostly in the hippocampus. This tracks with the memory functions (heavily concentrated in the hippocampus) helping to keep the question in mind as the trial went forth. On the whole, we are seeing "binding"- that mythical beast in the philosophy of consciousness, and of mental processes more generally.

• Universities should have nothing to do with business schools.
• What is our political system about, after all?
• Report number 9.
• Triple win from the Norwegian wealth tax.

Another Eukaryotic Gene Regulator Turns up in Archaea

Leucine zipper proteins as gene regulators in Archaea, regulated by nutrient-binding PAS domains.

In this day of AI, it is a pleasure to read an elegantly constructed and written paper that breaks new ground in biology. A group in Berkeley has discovered that a common gene regulatory family in eukaryotes, the leucine zipper proteins, have clear antecedents that were impossible to detect with pure sequence comparisons, but which turned up during their work on the nutritional regulation in archaeal methanogens. They note that archaea are now well-known to be the progenitors of the nucleus and nuclear functions of eukaryotes, (as well as translation, the cytoskeleton and much else), but that there has been a shortage of known regulators to account for the complex lifestyles of archaeal cells, particularly these methanogens.

Methanogens, as one might guess, give off methane, creating small amounts of the natural gas we take to be a mineral resource. For them, methane is an electron dump, (instead of oxygen, as it is for us), from food sources such as methanol, methylamine, and many other methylated compounds, which are slightly digested to form CO2 and methane. Methanogens are anaerobic, meaning they are poisoned by oxygen and originated from the early epochs of earth history before there was much free oxygen. It is naturally critically important for these organisms to know what kind of food is out there, since digestion of methylamine calls for different enzymes than the digestion of methanol. The authors thus had the hypothesis that there should be some regulatory apparatus for sensing methylamine (in this case) and activating the right genes to digest it. 

These authors performed a laborious laboratory evolution experiment, looking for mutant bacteria which were able to switch between food sources faster than the wild type Methanosarcina acetivorans, which take about four days to switch from methanol to trimethylamine. There were replicate populations, repeated passages of these slow-growing bacteria to new cultures with different food sources, (all under anaerobic conditions), and sequencing of the resulting strains. And then CRISPER-based specific mutagenesis to confirm the mutations that caused the phenotype of more rapid food switching. In the end, they obtained strains that took only 17 hours to switch food sources- much faster than the wild-type. The most common gene that was mutated was MA2561, which is a repressor of a methyltransferase specific to trimethylamine. These mutations disable the repressor and allow this gene to be expressed all the time, which would be expensive for the archaea in the wild, but in the lab, dramatically shortens the time needed to acclimate to the new food source. They renamed this gene amzR, for amazing regulator!, er, actually ... "archaeal metabolite-sensing zipper-like regulator".

The interesting thing is that this regulator and its family had not been recognized in archaea or in this species previously, even with all the computational power available to find homologs of any kind of protein. But once it was found by this functional screen, and the authors paid some attention to it, and put it through the alpha-fold program, then everything became clear. It is structurally similar to a large family of eukaryotic DNA-binding gene regulators called leucine zipper or b-zip proteins. This is because they contain a dimer of stiff alpha helices that wind around each other (using hydrophobic leucines at key contact positions) and make a scissor-like fork whose (basic) end binds to DNA. 

Comparison of leucine zipper structures. The authors provide an overview of structures. A is a garden-variety eukarotic b-zip protein, (GCN4), a regulator of amino acid biosynthetic genes. B is a more complex basic semi-zipper protein that combines DNA binding with PAS domains that are, in this case, sensors for oxygen levels, with significant psychiatric implications. C shows the newly found amzR repressor, in full dimeric mode with its PAS and GAF domains, which appear to sense trimethylamine. Lastly, D shows a helix-turn-helix DNA regulator that is characteristic of bacteria, though common in archaea and eukaryotes as well.

In addition, amzR has a couple of other domains- a PAS domain and a GAF domain, which are similar to each other, and are common motifs that bind small molecules, in this case likely the trimethylamine that causes the whole chain of events. The authors show that amzR binds DNA, that it represses the MT1 gene that allows growth in trimethylamine, and that it binds to trimethylamine itself, which impairs DNA-binding, presumably through structural shifts in dimerization or the orientation of the dimer. Thus, it is what is quite common in archaea and bacteria- the one-component-regulator which both senses and effectuates some kind of control in response to environmental conditions. 

A somewhat indirect, but effective, demonstration of ligand binding. The amzR protein melts at high temperature, here tracked by its fluorescence. The melting temperature shifts significantly upward if a (presumably stabilizing) ligand is offered to bind. NS = no substrate; MMA = monomethylamine; DMA = dimethylamine; TMA = trimethylamine.

They also find, using the new sequence, that direct homologs of amzR are common among methanogen archaea, and speculate that more distant relatives may account for some fair portion of the missing regulatory capacity that was remarked upon above. Unfortunately, given the inability of sequence comparisons alone to find these homologs, given the deep divergence in time among all these organisms, it remains difficult to speculate just how widespread this regulatory motif is in methanogens, or in archaea generally. But this work gives some hope that other whole families of regulators remain to be found, filling out what is obviously a complex regulatory apparatus in these organisms. Alpha-fold and other AI tools will likely be helpful in that quest.

The authors offer a basic model of control, where amzR (aka MA2561) binds to and represses the multi-gene operons controlling trimethylamine digesting methyltransferases among others. The inducing chemical, TMA, binds to and inactivates this repressor, turning on the operons.

• Thoughts on retirement savings. But honestly, we could blame the trillionaires and billionaires as well.

Towards a More Perfect Genome

The human reference genome is still evolving and improving. 

The original human genome, when issued to great fanfare in 2000, was an incomplete draft, covering only 92% of the sequence. It caught the most important parts, surely- most of the protein-coding genes- but sequencing technology was not up to the task of doing a complete job. To understand why, imagine the genome as a jigsaw puzzle. The number of pieces corresponds to the sequencing technology- the number of consecutive nucleotides (nt) that can read from the genome at one go. The dominant technology then (and still now) is short read sequencing, which reads about 200 nt at a time, from a randomly sheared up pile of DNA from the genome. Perhaps 300 or 400 if lucky. This means that the 3 billion nucleotide genome is chunked into ~15 million pieces, and that our one-foot jigsaw puzzle has 4,000 pieces per side, each one about 0.08 mm on a side. That is very small. Imagine trying to put together a puzzle like that! Computers are helpful, but they can only do so much.

And like a really hard jigsaw puzzle, our genome has large areas of uniform color- those sky-blue regions and large clouds. Those are the repetitive sequences, which, all together, make up about half of the genome. These are mostly defunct transposons, but also pure repetitive DNA that makes up structural features like centromeres and telomeres. For instance, the sequence GAATGn (where "n" stands for any nucleotide), goes on repetitively for 28 million nucleotides on chromosome 9. That is not an easy puzzle to solve, using tiny sequence reads, no matter how much computer power you have.

A map of some of the repetitive areas of the genome, focusing on centromeres. Panel A gives a color code for different kinds of repetitive sequence. Panel C gives compositions of the centromere of each chromosome, which are highly variable and different in length as well. Centromeres are underlain by piles of genomic junk, really, with a few key functional sequences.

So the original project did what it could and released the draft, which was both a huge accomplishment and a large boon to biology ever since. This draft has been patched up several times with improvements. In 2022, finally, a complete genome emerged, which advertised itself as end-to-end, or "T2T", for telomere to telomere. This genome still has a gap on the Y chromosome, which is a degenerate mess. But otherwise, it fills in all gaps, brings in a hundred more protein-coding genes, two thousand other genes, and fixes innumerable other sequencing errors. What got us to this point?

The new technology was long-read sequencing, which is an amazing advance based on threading a single DNA molecule through a tiny pore (a nanopore, which is actually a bacterial protein). You can read the electrical resistance as each base passes through. Each base is ever-so-slightly different electrically, making it possible to figure out the sequence, step-by-step. These methods have been refined to the point that they can read 100,000 nucleotides routinely, and even attain four million nucleotides at a single go. That is revolutionary for closing difficult, repetitive genome gaps.

Reading sequence, sequentially, as it passes through a tiny pore.

This long-read nanopore sequencing also benefits from using relatively unprocessed, native molecules (not extensively amplified/copied) helps to reduce errors, though this method has its own error problems, mostly due to the fact that taking electrical readings off the pore is extremely sensitive and tricky. It is also hard to scale up, mostly due to the slow process of threading DNA through those pores, so long-read sequencing has not challenged the huge bulk sequencers for lowest-cost or greatest scale. But when you have to finish a sequence, one that is problematic, with low complexity, long-read sequencing is perfect. 

Lastly, what is all that junk DNA doing in our genomes? Mostly, it is honest-to-goodness junk, just carried along out of inertia and lack of selective attention (though there is the story of piRNAs). Transposons can occasionally re-activate and create new mutations, but there are way too many of them to get rid of. They also can gain significant functions, giving rise to new, functional genes and regulatory controls over other genes. So they are a sort of a junk heap / pick-n-pull that can be very useful from time to time. The satellite sequences in the image above are concentrated in structural areas like the centromere, where they do not have classical "gene" type functions, but facilitate equally important processes like genome division and distribution to new cells. 

Unfortunately, as we learned from the draft genome, it is not a holy grail full of secrets that will, of themselves, solve our medical mysteries and confer immortality. Its use is as an armature on which we can hang the accumulating knowledge of biology. Each nucleotide in the genome has its own story, of how it got there, what processes it plays in, and how it can affect our health and happiness. The system built upon the sequence is far more involved and interesting than just the sequence.

• More on taxes and inequality.
• Who is Mr. Kremlev?

Sensing Unfolded Proteins

Do cells take joy in tidying? Absolutely! But how can they tell where the messes are?

Any household knows that without cleaning and disposal, nothing else is possible. Things pile up, messes accumulate, everything comes to a standstill. Cells are the same, needing a constant flow of new materials in, trash out, recycling, and cleanups. One of the major kinds of mess in cells is unfolded proteins, which when they accumulate can cause aggregations and diseases like Huntington's, other dementias, atherosclerosis, and inflammation generally. While there are a lot of mechanisms (chaperones, chaperonin cage complexes, proteasomes) dedicated to proper folding, sometimes they are not enough, and excess junk in the form of unfolded proteins builds up. One major response this is the unfolded protein response, or UPR, which is centered at the endoplasmic reticulum (ER).

Why the ER? This is where countless ribosomes dock to feed in nascent proteins destined for secretion, the plasma membrane, and many locations inside the cell other than the cytoplasm. So it is a key organelle for protein synthesis and distribution, and thus for sensing how cellular proteins are doing. This sensing, the UPR, starts with a sensor protein, called IRE1 (also called ERN1), which sits on the inside face of the endoplasmic membrane. IRE1 binds to unfolded proteins inside the ER, and when the concentration rises high enough, it gathers into condensates of its own that trigger a variety of responses that include, in the extreme, cell death. Yes, lack of timely tidying can have serious consequences!

IRE1 is multifunctional, containing a kinase that activates AMPK and JNK, proteins which influence metabolism and inflammation. It also contains an RNAse enzymatic activity, which activates a whole other program by altering the mRNA of XBP1, which then becomes active and produces a transcription regulator that starts building more ER, more chaperones, and more protein folding capacity generally. The RNAse also digests other mRNAs indiscriminately, lightening the load of translation and thus incoming nascent proteins. So IRE1 is a key homeostatic regulator that adjusts the size of the ER to cellular needs, and informs the rest of the cell how things are going in the ER. 

One structure of IRE1, luminal domain only. This is a dimer, with halves shown symmetrically across the middle. How does this bind unfolded proteins?

But how does IRE1 sense unfolded proteins? A recent paper went through a modeling study to show how IRE1 operates. Sitting in the ER membrane, the enzymatic, signaling parts of the protein are outside, while the sensor is inside, called the core luminal domain (cLD). The authors demonstrate that this domain binds unfolded proteins directly, using them as bridging elements to bring multiple IRE1 proteins together and form the glob that nudges aside the inhibitory protein BiP and promotes the phosphorylation activity and thus activation, of IRE1. BiP also binds to unfolded proteins itself, so forms another sensor that indirectly activates IRE1 under stress conditions. 

Firstly, IRE1 is a dimer, and the luminal domain is where it dimerizes. The structure (just of the luminal domain, above) shows a beta ribbon core and peripheral unstructured areas and alpha helices. After running the crystal structure (shown) though very brief molecular dynamics simulations, the unstructured areas flop around dramatically, while the core and dimeric structure remain stable. Intriguingly, the beta stranded center of the dimer resembles (between the blue helices, above) the peptide binding cleft of MHC molecules- those that present random antigens to T-cell receptors of the immune system. Yet the authors show that unfolded proteins bind differently to IRE1. They flop across the whole central region, attracted by negative charge as well as the open hydrophobic surface. The next image shows molecular dynamics before/after shots of various unfolded peptides known to bind IRE1 in vitro, showing how after starting straight across the central region, they each find different comfort zones, mostly over this central region, but some off to a side. 

Collection of IRE1 structures (gray) with unstructured peptides of various kinds bound to the luminal face. t=0 is the starting point for the molecular dynamics simulation, which went for one microsecond. End points are below. In each case, the unfolded peptide stayed bound, and settled into some comfortable position, based on the hydrophobic and negatively charged landscape of the IRE1 face. 

The intriguing thing about this binding is that the structure of IRE1 itself is unaffected. The dimer interface remains stable and open, leading to the hypothesis that these peptides form bridges by becoming the meat of a two-dimer IRE1 sandwich. And it is this that then displaces the inhibitory protein BiP and leads to condensation and activation, by bringing together multiple IRE1 kinase domains (on the outside of the ER), and activating their kinase/signaling functions, turning on the various downstream pathways.

A helpful model from another lab, describing how, once the unfolded proteins are bound inside the ER (by mechanisms unknown at the time), the outside domains would gather, trans-phosphorylate, and activate downstream pathways. 

• Those flock cameras.
• As if charging for previews of presidential posts wasn't corrupt enough.

Our Loopy Way of Learning

Different ways of learning go through different parts of the brain.

Humans are champion learners. Other animals are smart, but we are smarter, spending more time in childhood soaking up the mysteries of the world around us, and storing them in a larger and better brain. Learning is our calling card, allowing us to adapt to any environment and defeat any foe. Indeed, we have overwhelmed the earth's biosphere, and need to exercise some deeper and longer vision by pulling back from our successes in mastering every possible resource.

But how the brain does all this is something we have yet to fully learn about. Our sense organs bring in tons of information, but as we have learned in computer science and AI, it takes a lot more to create usable information than just amassing data. How do the various parts of our brains work together to create actionable models of the world out of that data? An innovative theory was elaborated a quarter century ago by Kenji Doya, who has apparently gone on to a research sideline on soccer(!)

The problem began with, in part- what does the cerebellum do? From lesion data, it has long been clear that this small part of our brain has a big role in motor accuracy and coordination. But cerebellar neurons project all over the brain, not just to motor areas, and it gradually became clear that the cerebellum affects those other areas as well, helping us learn and manage many cognitive tasks. But structurally, the cerebellum is a peculiar organ, highly repetitive, with relatively linear and parallel processing, a bit like a GPU. It is clearly specialized for something- some kind of fine tuning, but what?

Doya's theory about the distinct learning styles contributed by different parts of the brain.

Doya proposed a general theory about learning in the brain, which splits learning up into three types. One type is supervised learning, which is what the cerebellum facilitates. We grab something, we recognize an error in eye-hand coordination, and we learn to grasp better. There is a loop involved, both physically and logically, by which some task is evaluated, and error signal sent back to the source, and a small adjustment is made for the next iteration. Humans have great hand coordination, partly thanks to our advanced cerebellar-cortical connections. 

The second form of learning is reward-based learning. This prototypically features dopamine, and is centered in the basal ganglia, where emotions are processed, but which is also a major switchboard for connections upwards to the cortex and downwards to the brainstem, and is also critical for motion control. Parkinson's and Huntington's diseases both affect this region. Doya proposed in broad terms that this area doesn't do the kind of error correction that the cerebellum does, but rather learns by reward- whether the action's goal was reached. If the goal was reached, a spurt of dopamine is issued, which reinforces whatever connections helped that action happen.

Interestingly, there is a progressive temporal aspect to this learning. As an action is learned, the reward happens sooner, given the predictive capability of our learning / cognitive systems. The whole point, after all, is to figure out as early as possible what will be happening in the future. If a bell is associated with food, at first the food prompts the reward. But as learning happens, and the bell is reliably associated with the later food, and the bell starts to set off the reward all by itself. Gradually, the sight of the person getting about to ring the bell, or the expected time of day, sets off the reward... it is cast progressively back to whatever the perceived cause is, going back in the learned causal chain. Eventually, as everything becomes rote and the novelty wears off, reward lessens, and learning is complete. Mealtime is here, as it is every day... boring.

Lastly, the third type of learning is unsupervised learning. This is the job of the cortex, and is the most interesting and subtle form of learning. The world has patterns, and the cortex's job is to figure out what those patterns are. No one outside tells it what is right or wrong, what is or is not real. It has to figure all this out from the stream of input, including its own blundering actions and experiments. Unsupervised learning is thought to be well suited to the basic mechanism of the brain, its Hebbian circuitry where activity reinforces any active connection. There are a lot of mathematical methods that have been devised to approximate such processes, like principal component analysis. Data can be binned / categorized in various ways to simplify and organize reality, and eventually we have ... language and concepts, labels and abstract understanding. This categorization of the world is immensely powerful, if always inaccurate and approximate. 

Doya's model of the cerebellum. Purkinje cells at the center are also conceptually at the center, as the only cells that are plastic in this scheme, able to change their connection weights due to input from the other layers, which in turn get input from other brain areas including the cortex and thalamus. This has been likened to a perceptron in computer science terms.

Doya thus created an influential heuristic that divided up learning styles / mechanisms by anatomy in the brain. Obviously, all these mechanisms work closely with each other. The cerebellum doesn't know anything without conceptual categorizations coming from the cortex, which provide the foundation of error signals. Similarly, whether goals have or haven't been reached has to be informed by conceptual data from the cortex (or food signals, or other sensations from elsewhere in the body). Learning is a whole-brain activity, and while it is prodigious and effortless early in life, it continues to be essential throughout, lest we be overwhelmed by the fresh challenges the world keeps presenting to us.

• Recent work fleshing out this theory.
• Thought crime.
• Empathy as crime.
• .. or as being Democratic.
• Scrap all fossil cars now.
• China is going after their rich.

Ohnologs and Paralogs: The Wages of Gene Duplication

Two whole genome duplications lie at the root of vertebrate evolution.

Another week, another story about the power of gene duplication in evolution. While rare during normal reproduction, gene duplication happens pretty frequently over longer time scales. How else would we get a thousand olfactory receptor genes, all similar to each other? But other accidents can occur as well, like whole chromosome duplications (such as what leads to Down syndrome), and whole genome duplication, when the cell division process stops early, but otherwise proceeds with cells remaining viable with double the genomes as before. This is common in plants. Corn is tetraploid, wheat is hexaploid, and strawberries are octaploid. 

But among animals, whole genome duplication is less common. Two decades ago, however, two researchers working from the newly sequenced human genome revealed that there were two such duplication events at the beginning of vertebrate evolution, explaining some oddities and also perhaps the speed and power of subsequent evolution. The basic evidence is the genome sequence, which is full of related genes. Genes that do the same thing in various species, and are lineally related, are called orthologs. That is relatively simple, per the Darwinian tree of descent of all life. Genes within one organism / one genome that are similar to each other due to ancient duplication events are called paralogs. All those olfactory receptors are paralogs, for instance. Lastly, genes that are paralogs stemming from a whole-genome duplication event are called ohnologs, in honor of Susumu Ohno, who led the field of molecular evolution in recognizing the importance of gene duplication, and speculated about whole genome duplication well before it was discovered.

A classic example of this evidence is the hox cluster, a linear sequence of genes that have been extensively studied in flies as providing an important set of regulatory controls over the linear body plan. They lie in the middle of the developmental cascade, downstream of egg and body polarity genes, but upstream of specific appendage and tissue expression programs. They encode DNA-binding (homeobox) transcription regulators, and their position in the genome is co-linear with the body parts they activate because there is a progressive chromatin opening process by which this whole locus becomes activated. Well, flies have one hox locus, but vertebrates have four. What happened?

Hox loci across evolution. Where flies and primitive chordates have one hox cluster, vertebrates have four. How did that happen?

Obviously, once researchers lined everything up, it became pretty clear that there were two massive duplications along the way, creating four hox loci in the genome, after which quite a few of the duplicated genes fell away. After a duplication event, gene survival is a race between neo-functionalization (which leads to preservation by selection) and deleterious mutation, degradation, and disposal. Enough of these ohnologs survived to help fuel the substantially greater complexity of the vertebrate body plan, now including intricate wrists and hands, and ever more involved head structures.

Similar findings were made all over the human genome. The original paper has a graph that shows that, across the genome, most paralogs exist in families of four. If genome duplication were not the applicable hypothesis, then one would expect a smooth asymptotic curve downwards from one member (implicit) to two, then three, and fewer from there outwards, since single gene duplications would each be independent statistical events. But no, there is a peak at four, indicating that some process yoked together many, many genes into parallel duplication events, twice in succession. 

A graph of count of paralogs vs their frequency by count, in the human genome. What should in principle have been a smoothly declining curve from 1 or 2 turned out to have a weird peak at the number four. This was important evidence that many of our paralogs arose through some common event, such as a pair of ancient whole genome duplications. 

OK, so far, so old hat. A bunch of more recent papers flesh out this story a bit, showing how the vertebrate duplications were timed, and how they affected various types of genes. miRNAs, for instance, turn out to be ancient genetic elements, and were duplicated along with everything else. miRNAs that originate from (and were preserved from) the whole genome duplications tend to be more conserved, have more targets than other miRNAs, are more highly expressed, and have higher rates of targeting RNAs of transcription regulatory proteins that likewise originated from whole-genome duplication. The traces of this history are thus interestingly preserved.

Another paper traced the effects of the vertebrate ohnologs on brain development. Using the pre-vertebrate amphioxus as an out-group for comparison, they find that the ohnologs that date from the whole duplications are more highly associated with brain cell types and their developmental programs than are other gene duplications- either ones dating from the same time as the whole genome duplications, or since.

"Compared to their closest invertebrate relatives—tunicates and amphioxus—vertebrate brains are highly regionalized and complex."

"Ohnologues were enriched in development, cell-fate commitment, signaling and neurotransmitter transport. By contrast, SSD [small-scale duplication] paralogues were enriched for immune response and sensory perception in all species, a result that matched previous reports."

Lastly, a paper focusing on intermediate genomes, part of a plethora of genome sequencing that has happened since the original analysis, reveals exactly what happened at this time, about 500 million years ago. From a molecular perspective, hagfish are in the same group as lampreys- both parasitic eels that lack real jaws but arise from the vertebrate lineage. They are cyclostomes, while we are gnathostomes. After the first whole genome duplication about 530 million years ago in the common stem lineage, the gnathostomes and the cyclostomes split and each experienced their own, separate genome duplications roughly 490 million years ago. This can be concluded from the differing gene collections that survived from each respective duplication, and their sequences relative to the cyclostome/gnathostome split. Indeed, lampreys and hagfish have six hox clusters instead of four, indicating that a triplication event happened, instead of duplication event, between their split from gnathostomes and the split between lamprey and hagfish. 

A phylogenetic tree with dates, locating the genome duplications deep in the vertebrate stem lineages. 1R is the first genome duplications, common to all vertebrates. 2R is the duplication in the lineage leading to jawed vertebrates, while CR is the apparent triplication that happened in the lineage leading to cyclostomes. Apologies for the antiquated human being. 

All this genome renovation didn't benefit cyclostomes the way it did gnathostomes, however. Only one group went on to globe-straddling glory, indicating that while genome duplications may be helpful for evolutionary / developmental innovation, they certainly aren't determinative. They are grist for an evolutionary process that remains largely shrouded in mystery- the context and environments of the time, the competitive biosphere, and the internal molecular environment. There is no reason to think that all this is unaccountable by natural processes, but that doesn't mean we have the information to reconstruct it in detail. We should thus be deeply grateful that the scientific community can dig up even this much of our history.

• But how can we take more money away from workers?
• The smearing of Anthony Fauci. And of basic reason.
• Reward and addiction.
• Still stupid ... the Discovery Institute and "information".

Revealing the Origin of Eukaryotes

Tracing the breadcrumbs of deep phylogeny through gene duplications. 

Last week, we discussed the jamboree of mutation that is the MHC locus in animals. One part of that story was the persistent duplication and pseudogene-i-zation (which is to say, the birth and death) of genes in that locus through primate evolution, another consequence of the never-ending arms race with our pathogens. Gene duplication has much deeper significance, however, as it has given rise to new genes throughout biological evolution, as it is merely an accident of DNA (or RNA) replication, thus operationally rather frequent. Significant transitions, such as the advent of eukaryotes, and the advent of plants and of angiosperms, all feature rapid gene duplication, even whole-genome duplication, which provide grist for new functions, refinement of old functions, and speciation.

A recent paper did a deep dive into the origin of eukaryotes, perhaps the watershed event in the history of life, second only to the origin of life itself on the early Earth. It is very difficult to reconstruct what happened during these extremely early events, which these authors put as long as three billion years ago. There has been so much mutational churn in our molecules, and no fossil record to speak of, that, again like the origin of life itself, we do not have a great deal to go on. Thankfully, the discovery of the kingdom of archaea, and its especially eukaryotic-like class, the Asgard archaea, has provided a much better platform for speculating about what the originating organisms might have been like.

Eukaryotes differ from bacteria and archaea by numerous characteristics, implying extensive development/evolution before the last common ancestor from which we can trace all the extant descendants. These include the eponymous nucleus, an internal system of membranes and membrane-bound organelles, a complex cytoskeleton of both actin and microtubules, a mitochondrion, meiosis, a microtubule spindle-driven division system, and many other, more obscure molecular differences. The lack of intermediate forms is certainly frustrating from a scientific perspective, making it difficult to go through the catalog of life today to identify the steps involved in this transition. It also implies that the process ended with an evolutionary bang that in essence caused the final iteration to wipe out all the preceding forms, a bit like humans vs the many other Australopithecines and their descendants. 

These differences are enormous and momentous, making eukaryotic cells far larger, energetically capable, and complex than their antecedents. Another key characteristic is a prevalence of gene duplication and specialization, leading to much larger genomes along with all the other complexity. It has long been theorized that it was the union of the proto-mitochondrion, which was a bacterium, with the archaeal founding cell, that set the whole process off, particularly providing the massive amount of energy required for all this rise in complexity. 

However, the current authors claim that this definitively is not the case- that the mitochondrial symbiosis was a rather late event, coincident with the oxygenation of the atmosphere about two billion years ago. They argue that, since gene duplication and specialization is in any case a clear characteristic of eukaryotes, using molecular clocks to time these many gene duplications can give us relatively specific and data-rich insight into the epochs at which each of the processes that they participate in first developed. 

The authors present estimated times of origin for key duplications involved in RNA synthesis. The RNA polymerases I, II, and III were duplicated from one polymerase in archaea, and are composed of numerous gene products, as well as ancillary regulatory proteins. While some subunits are still shared, the ones that duplicated did so in the range of 2.3 to 2.8 billion years ago, before the symbiosis with the mitochondrial precursor, timed at "mFECA", or the mitochondrial first eukaryotic common ancestor, about 2.2 billion years ago. "nFECA" is nuclear first eukaryotic common ancestor, while "LECA" is last eukaryotic common ancestor. The axis at the top is time before present, in giga-years ago, or Ga.

For example, eukaryotes all have three RNA polymerases, while bacteria and archaea only have one. These RNA polymerases specialize on mRNA synthesis (RNA pol II), ribosomal RNA synthesis (RNA pol I), and tRNA, 5S RNA and U6 snRNA (RNA pol III). Why this specialization exists remains a bit mysterious, though once it started there was no going back. In any case, it happened and is totally characteristic of eukaryotes, so these duplications (which involve the duplication of numerous genes encoding proteins of the polymerase complex itself and its regulators and way-finding helpers) can be used to date eukaryogenesis, using a very carefully calibrated and species-sampled molecular clock. And what they find is that all of these duplications, given as ranges in the graphs above, with green dots at the likeliest time points, all lie prior to the point of "mFECA", which is the mitochondrial first eukaryotic common ancestor, at roughly 2.2 billion years ago. Tabulated over studies of almost a hundred other genes that are similarly duplicated and characteristic of eukaryotes, the authors place the "nFECA", or nuclear eukaryotic common ancestor way back at almost 3 billion years ago. 

That is a very long time! It is a very long time ago that the characteristics of eukaryotes started to come together, and it is also a very long time (well over a billion years) over which that development spanned before culminating in the last common ancestor of all extant eukaryotes, which the authors put at about 1.7 billion years ago. And note that that still comes a billion years before the Cambrian explosion of animal life. We are talking about really deep time here. 

If mitochondrial symbiosis came later, (causing another bout of genomic change as hundreds of bacterial genes moved from the symbiont to the nucleus), just as the atmosphere was getting oxygenated, what was going on in the preceding billion years? One can speculate that these evolving proto-eukaryotes were the top predators of their world, a bit like eukaryotic protists in microbial environments today. The competition would as always have been intense, but there were no higher life forms (i.e. animals or protists) to worry about. Size was not a problem, but rather an advantage. The focus was on quality and effectiveness in finding, disabling, and digesting prey. Thus there was little downside to complexity, unlike the constraints on prey species, which had to optimize for rapid growth and reproduction under both nutrient and predation constraints. New and costly systems for protein secretion, for prey digestion, and intelligent, responsive regulation of all cell processes might have been consistently advantageous.

I take this work as being definitive, at least in terms of settling the debate about early vs late symbiosis. It benefits from a clear theoretical basis and new, voluminous sequence data from appropriately related species in its molecular comparisons. I recommend it highly, as I can only scratch the surface of its analysis here. The advent of mitochondria was then a later step change in complexity, but brought incredible metabolic productivity that, at least in oxygen-rich environments, would have definitively superseded all the prior forms of the proto-eukaryote, and (unfortunately for us) erased that record of evolution. A later innovation, placed between mFECA and the last eukaryotic common ancestor "LECA" was sexual meiosis, another fascinating and characteristic innovation of eukaryotes. These authors thus provide a true revolution in our understanding of eukaryogenesis, extending it across well over a billion years.

One more sample from the author's presentation of molecular duplications that help time eukaryogenesis. Here, DNA repair proteins are emphasized, including some involved in meiosis (light blue dots, see legend). Some of these duplications involve genes from the mitochondrial endosymbiont, (MSH4), and some (such as EME1/2 and MUS81) come from older repair processes, but later specialized for meiosis. 

• Critique of our court.
• A vision .. "in which elections are hollow exercises"
• The end of structure, the end of truth.
• A confluence of corruption, bad faith, and poop.

MHC Through Evolution: Breaking All the Rules

The immunologically critical MHC gene cluster plays by its own rules through the arms race of life.

Last week, I discussed the very general landscape of variation in the human genome, specifically the tradeoff between prevalence in the population and effect size. Given that the vast majority of variants are deleterious, those that affect our phenotypic traits are more heavily selected against the greater effect they have. The result is that at a gross level, most genes and most traits share the same general distribution of lots of variants (alleles) with minor effects, and far fewer with large effects. And those with minor effects also turn out to be tangential for biologists, rarely informative about the nature of the traits they (sort-of) affect.

This week, another paper and another view of evolution, though the eyes of one the more critical genes of the immune system, the major histocompatibility complex, or MHC group of genes. While our adaptive immune system has developed the extraordinary and powerful ability (though semi-controlled DNA recombination of the antibody and TCR genes) to recognize practically any antigen, foreign or domestic, that system requires stringent controls. One of those controls on T cells, which carry the antigen-recognizing TCR receptor, is that it can only see antigens that are "presented" on MHC molecules. MHC proteins have a surface cleft that gets loaded with and holds small peptides (8 to 12 amino acids long) that are cleaved from other proteins, either from pathogens or from the cell itself. The MHC+peptide complex then sits on the surface of the cell, announcing either that 1) I am healthy, full of normal cell proteins, going about their business, nevermind, or 2) I have some other proteins inside, either from a viral infection (class I MHC) or from some bacterium I have just phagocytosed to deal with an infection (class II MHC). In the second case, T cells carrying the TCR receptor lock onto the MHC+peptide complex, and start up the process of killing that cell. 

MHC proteins (beige, pink) present an antigen (red) from one cell, and dock to a T-cell which recognizes the antigen+MHC complex by shape, using the TCR receptor. The additional CD8 or CD4 receptors help to verify the proper binding. While class I MHC are present on all cells, class II MHC are on phagocytosing cells like dendritic cells and macrophages that commonly ingest, reprocess, and re-display bits of encountered pathogens.  

Given that the TCR gene recombination process is unbiassed and produces a galaxy of random binding specificities, how do these cells distinguish self from non-self antigens? This is a deep question that is not fully resolved. But one major mechanism is thymic selection, which is what gives T cells their name. Special cells in the thymus display a wide range of self-antigens, and T cells, which are obliged to pass through the thymus during their maturation, are induced to commit suicide if they react to any of them. A paper from 2018 fascinatingly discussed how it is possible to create a T cell population that knows the "language" of foreign vs domestic after what is known to be a rather haphazard selection process, which displays only a partial range of self-antigens, and leaves quite a few self-reactive T cells around.

At any rate, the MHC proteins do not benefit from hyper-variation provided by genetic recombination. Yet it turns out that variation is beneficial here as well. The way foreign antigen peptides nestle in the MHC groove can be varied by mutations in the MHC molecule, providing a rich field of variation in antigen recognition and thus disease resistance. So, our MHC genes have not just a few alleles in the population, not just a few dozen, but over six thousand alleles. For each individual MHC protein, each person has only two, but over the population, there myriads with different properties. MHC was first recognized for its role in self vs non-self recognition and transplant rejection, (thus the "compatibility" in its name), and it quickly became evident that people vary tremendously in their MHC complement. And this variation plays a big role in keeping us (and all other animals) going as populations, in the face of pathogens that evolve a lot faster than we do. 

A recent paper provided a phylogenetic history of MHC molecules in monkeys, covering the last sixty million years of evolution in our lineage. It is a festival of gene birth, death, and duplication, quite apart from the smaller mutations that are constantly accumulating and cycling through the population. The MHC region carries about 200 related genes, most of which have minor roles, and only six of which (three MHC class I, and three MHC class II) follow the high-mutation pattern because they encode the main antigen presenting proteins. These genes are subject to, quite obviously, unique selective forces. 

How the MHC gene cluster looks, when aligned and identified by gene, over the primates. Note the deep divergence between the new- and old-world primates. Genes A, B, and C are the major MHC class I genes, which vary the most over this time. Note also how some of these genes have gone through extensive duplication in some old-world monkey lineages.

The first force is balancing selection. As soon as one allele becomes common, pathogens evolve to evade its presentation skills, rendering it less effective and less desirable. This results in a population full of minor variants. Indeed, for any individual person, having two MHC molecules that are the same would be bad. Being heterozygous at these genes is highly advantageous, thus enforcing both the retention of minor alleles, and an observed behavior in mating to favor partners with different MHC complements. Apparently, our MHC makeup is reflected in our personal aroma! 

A second force, conversely, is the retention of ancient alleles. It turns out that, across the old-world monkeys, many MHC alleles are preserved and cluster more closely in sequence comparisons with each other than they do with other alleles in the same species. That is, despite the general speed of MHC evolution and constant accumulation of new alleles, old alleles are preserved in all monkey populations as well, due to their distinct capabilities, under balancing selection. This is part of what makes population bottlenecks so damaging to near-extinction species. They lose critically valuable genetic resources (in the form of rare MHC alleles) that represent millions of years of accumulated variation. 

Incidentally, the trees shown here again reinforce the history of primate evolution, with new world monkeys splitting off from the old-world monkeys quite early on and developing a very distinct set of MHC molecules.

So, while virtually every other gene in the genome is being relentlessly optimized, sticking to its knitting, doing one thing and being beaten down whenever any mutation steers it from its optimized path, the MHC genes follow quite a different path, at least in portions of their sequence that provide variation in antigen binding and presentation. These genes revel in endless diversity, throw off pseudogenes at a high rate, wink out of existence and come back in other forms. Natural selection is the motor in each case, but meets the challenge of survival in different ways.

• We are still OK.
• But the climate.. is not.
• Outstanding overview of China and Xi Jinping thought.
• A total empathy deficit.