Field of Science

Wait, there's a much simpler explanation! (For CRISPR-Cas, not for GTA)

I'm in Halifax for a couple of weeks, visiting Ford Doolittle and his philosophical colleagues,  We've spent much of the time considering the extent to which CRISPR-Cas systems can or should be considered 'Lamarckian'.  I started with the simplistic perspective that of course it is, because an acquired character (immunity to future phage or plasmid infection) becomes inherited because the Cas proteins insert short phage- or plasmid-derived DNA sequences as a CRISPR 'spacer' into the chromosome.

Here's a very detailed diagram I made of the evolutionary events (mutation and selection) affecting CRISPR-Cas systems (click to embiggen):

We ended up concluding that 'directed mutation' was a better perspective.

But, once our ideas started settling down, this detailed diagram got me thinking about how uncertain and far in the future the 'immunity to future infection' benefit is.  That's a problem for CRISPR-Cas evolution, since this uncertainty greatly weakens the selection maintaining and refining the system.  Iv selection is too weak, the system shouldn't be maintained at all.

A more urgent problem is that the cell needs to survive the immediate infection/invasion before it has any chance of benefiting from the long-term immunity.  This becomes especially important if the bias against potentially-lethal self-spacers arises because the cell contains many copies of the invader genome.

But the cell does have a very nice mechanism to clear the invader, because it has just created an invader-specific spacer in its CRISPR array.  Transcribing this new spacer would give it many copies of an invader-specific crRNA with which Cas9 can destroy all the copies of the invader genome.

So here's my new hypothesis:
The primary function of CRISPR-Cas systems is the detection and immediate destruction of phage and /or plasmid DNA.  Benefits from immunity to future infection are relatively unimportant. 
Things I need to find out:

 Is this a new idea?  I don't remember seeing it anywhere, but if any reader knows of a prior proposal please let me know in the comments or via Twitter (@rosieredfield).

Is relevant data available?  The basic experiment is, in principle at least, quite simple.  Do cells with an intact CRISPR-Cas system survive phage infection better than cells with a defective system?  Do they become transformed less efficiently by plasmids?  These tests would be most sensitively done under sub-optimal infection conditions.

How is transcription of the Cas genes and CRISPR array regulated?  In particular, how efficiently is the CRISPR array transcribed and processed immediately after a new spacer has been added?  In the context of my GTA-as-CRISPR-vaccine ideas (see this post from a few months ago) I'd been looking for reports that new CRISPR spacers can be immediately transcribed, creating crRNAs that can immediately attack the original invader.  I didn't find any solid data, but neither did I find anything that ruled this out. 

R. capsulatus growth curves in RCV medium

My upstairs GTA colleague and I were surprised that the Bioscreen growth curves in the previous post didn't show a dip in OD600 of the GTA-overproducer strain like that seen in manual (non-automated) growth curves.  This dip is thought to be caused by the lysis of GTA-producing cells as GTA production peaks when cells hit stationary phase.

We thought part of the problem might be that I used the standard YPS medium which is based on modest concentrations of yeast extract and peptone.  The clearest/most-recent published demonstration that GTA-producing cultures used RCV, a simpler 'defined' medium based on malate, and showed that the apparent lysis occurred in medium with 0.5 mM PO4 but not in medium with 10 mM PO4. 

So I redid the growth curves for all 6 strains, using both high-P and low-P versions of RCV (kindly supplied by my upstairs colleague).  The results are not inconsistent with the Westbye results, but they're not at all compelling.  None of the strains decreases in OD600

The problem is that there's quite a bit of between-strain variation in growth and in the stability of the stationary phase OD.  (Within each strain the replicate wells give very similar results, with one exception.)

The graph below shows growth in the high-phosphate medium.  The main graph shows OD600 on a log scale (appropriate to exponential growth), and all the strains appear to stably reach similar densities.  But the inset shows the same data on a linear scale, which makes the variation look more significant.  The overproducer strain stops growing abruptly at OD600 = 0.7 a lower density than the other strains.

Here's the cells in the low-phosphate medium.  There's an initial drop in OD600, over the first 10 hours, but then all the strains grow steadily except strain YW1, where the individual wells grew at different rates for no apparent reason.  Again the linear-scale inset shows the substantial variation at stationary phase.  The overproducer DE442 again stops growing, this time at OD600 = 0.8, and now its OD falls by about 20% over the next 40 hours.

I really don't feel comfortable drawing any solid conclusions from this one experiment, especially since there's a blip in many of the growth curves at a point where I stopped and restarted the runs to add more time when I realized that 3 days wasn't going to be long enough.  Even though the shaking only stopped for 2-3 minutes, and the trays of cells remained in their holder with the lid closed, most of the strains had an abrupt change in OD600.  (You can see the blips at hour 63.)

Plan:  Do the run again.  This time I'll pre-grow the cells into log phase in high-P and low-P RCV.  medium (the upstairs colleague has offered me enough medium to do this).  And I'll plan on pausing the run at key times to take samples that I can assay for GTA production.

What can we learn from growth curves?

Here's the results of the Bioscreen growth curves I ran for Rhodobacter capsulatus strains:

Each dot is the mean OD600 of 15 replicate wells, each containing 300 µl of culture, with ODs read every 20 minutes for 45 hours.  The cultures all grew up at about the same times, but I've shifted the X-axes so the curves don't overlap.  OD values below about 0.015 are not significantly above the backround absorption of the culture medium. The Y-axis is a log scale, so when doubling time is constant the dots will fall in a straight line.

I did these runs 'just-in-case', because I'm going to be working with Rhodobacter capsulatus at Memorial University in Newfoundland for the next few months (on sabbatical leave) and thought they probably wouldn't have a convenient Bioscreen that I could use.

Now I need to figure out what we learn from these, and whether I should do any more experiments before I leave UBC.

The simplest expectation is that once the cells have adjusted to the medium (after 'lag phase') they will grow at a constant rate until they run out of nutrients or experience other bad consequences of high cell density (little oxygen, accumulation of toxic byproducts).  But all of these cultures instead exhibit 'diauxy', a mid-growth shift from one resource to another.  We see  this as a brief slowing or even cessation of growth at about OD=0.05 (orange shaded band), after which growth resumes, often at a different rate.  The pause occurs because the cells need time to adjust their metabolism to a change they've caused in the medium, such as exhaustion of one nutrient or new availability of another. 

I don't know enough about R. capsulatus metabolism to speculate about what the change might be, but it might affect production of Gene Transfer Agent particles.  The pause isn't due to lysis of GTA-producing cells, because it's not changed in the ∆∆ strains, which have deletions of the GTA gene cluster and lysis gene.

SB1003, B10 and YW1 are all 'wildtype' strains, I think.  Strain YW1 grows much slower than the others, although it still speeds up after the growth pause, and it reaches a slightly lower final density.

Strain DE442 carries a mutation that causes over-expression of the GTA genes and over-production of GTA particles.  Growth curves in a 2013 paper found that this strain had a substantial drop in OD once growth ceased, thought to be due to lytic release of GTA particles, but no drop is seen in the Bioscreen culture.  That work used a low-phosphate version of a different medium, RCV.  But an earlier paper found strong lysis with the same complex medium I used (YPS), and low lysis with the high-phosphate (10 mM) standard RCV medium.

The lab upstairs has both low-phosphate and high-phosphate versions of the RCV medium, so I'm going to repeat the time course with both.

growth time courses

In a few weeks I'll be headed for the Maritimes, for the final part of my sabbatical work on Gene Transfer Agent.  But before I leave here I want to run some detailed growth time courses on GTA-producing strains, taking advantage of the BioScreen machine belonging to the lab next door.

I'll first do a trial run with all the strains I have,  to check the basic growth kinetics under the Bioscreen growth conditions.  Then I'll see if I can combine the growth measurements with testing for the amounts of GTA produced.

Phage plaqueing still sucks - what to do now?

I feel like I've been sucked down a hole of trying to get consistently countable plaques from the Rhodobacter capsulatus phage I'm testing.  After seven weeks of plaqueing with various combinations of strains and agar concentrations and cell densities, I'm no closer to having a well-behaved phage I can use to test the GTA-as-vaccine hypothesis.

Along the way I've eliminated various sub-hypotheses:

1.  The plaques are tiny/faint/blurry/invisible because the phage capsids have long fibers that reduce diffusion through the top agar:  Test - use increasingly dilute top agar.  Top agar us usually 0.75% agar; I've taken this down to 0.3% (the lowest concentration that's still stable enough to handle). The first time I got somewhat larger plaques, but this was not reproducible.

2.  The plaques are tiny/faint/blurry/invisible because GTA gene products contribute to phage production:  Test:  Plaque phage on a GTA overproducer strain.  Result:  On the first try, plaques on the overproducer seemed larger.  But this was not reproducible.

3.  The plaques are tiny/faint/blurry/invisible because the GTA-as-vaccine hypothesis is true:  (Plaques can't grow because rapid diffusion of GTA particles allows surrounding cells to become CRISPR-resistant to the phage before the phage gets to them.)  Tested by plaqueing the phage on cells deleted for the entire GTA operon and for the separate endolysin.  Result:  Plaques on these '∆∆' strains are just as lousy (maybe more lousy) than on the GTA-producer parents.

4.  Variant (large) plaques contain mutations that increase infectivity or diffusion:  I made new lysates from a couple of big plaques that spontaneously appeared among the tiny plaques, but these lysates still gave tiny or no plaques

I know that the phage lysates do infect and kill the cells, and do produce progeny phages.   When I put a spot of sufficiently-concentrated phage onto a lawn, all the cells die, and when I make a lysate with this 'clear' top agar, I get way more phage then I put in.  Can I use the lysate to test the GTA-as-vaccine hypothesis even though I don't have countable plaques?

What would I do?  Here's an earlier blog post where I laid out a crude plan and a list of all the things I'd need to find out before actually doing the experiment that would test the hypothesis:

Luckily, after I wrote the above I made another grand attempt at titering the phages on the various strains.  Well, I made a sloppy attempt, learned from at and made a better attempt, which more-or-less worked. 

Basic test:  Pour lawns of the test strains, using cells concentrated from 400 µl of culture, in 1.5 ml of 0.4% top agar.  Put 10 µl spots of different dilutions of phage lysates onto these lawns, let the liquid absorb, and check the next day.  Yesterday I did this using photosynthetically grown cells (supposed to make better lawns) and today I've repeated it using cells grown aerobically in the dark.  Here's yesterday's result for one of the two phage and one of the six strains:

The central clear spot is undiluted lysate, and the other spots are 10-fold dilutions of that.  For undiluted, 10^-1, 10^-2 and 1-^-3, the spot is clear (all the cells have been lysed).  The 10^-4 spot still has patches of non-lysed lawn, and the 10^-5 and 10^-6 spots have distinguishable plaques.  Two of the four healthy strains gave countable plaques (27 and 29), which is nicely consistent.

I'll wait for tomorrow's results before proceeding.

Phage phrustration

Aacckk!  I've spent more than a month trying to get decent R. capsulatus phage plaques on R. capsulatus lawns.  Still no consistent success.  In one experiment I had much better plaques on cells of strain DE442 (a GTA overproducer), but that did not replicate.  I suspect that non-tiny plaques depend on exactly the right balance of the cells' physiological state, their density, the agar concentration, the culture medium, and other factors I haven't attempted to vary.

Yesterday I was almost ready to do a UV-irradiation experiment to generate mutant phages that make larger plaques, starting with two lysates I'd grown up from single plaques that were much larger than the rest.  But the dilution-series plates I was titering these lysates on grew up with very similar numbers of (mostly tiny) plaques, suggesting that phage contamination had crept into lysates with unexpectedly very low titers. 

This week I've gotten a couple of good suggestions from visitors:

The first visitor suggested that I give up on the characterized/sequenced phages I've been working with and just isolate a better-behaved phage from R. capsulatus's natural environment.  I'd have to learn how to do this (get water

Another visitor suggested that maybe the problem is the correctness of my hypothesis about GTA transduction of phage DNA leading to CRISPR-mediated phage immunity in the GTA recipient.  That is, maybe the first cells to get infected in my lawns produce so many phage-DNA-carrying GTA particles that many of the neighbouring cells that would otherwise be lysed by the phage become immune to the phage before the plaque can form.

There's a simple way to test this - see if the phage form better plaques on a strain that doesn't produce GTA.  So tomorrow I'm getting some GTA mutants from the guy upstairs.

Phage plans - let's put natural selection to work!

I have a two-pronged plan to get a phage strain that gives good enough plaques for my GTA-as-vaccine experiments.

I obtained reasonable titers of two phages, 'Titan' and 'Saxon'.  I'll invest a couple of weeks to see if I can get better and more reproducible plaques with either of these.  The genome sequences of these phages are not closely related.

First, improve the plaquing conditions:  The researcher who isolated the phages recommends using for the lawn cells that have been grown photosynthetically to a high density,  He also suggested trying a lower top-agar concentration.  I'll play around with these and other variables to see if I can get better plaques.

Second, use artificial selection to get a better strain of phage:  I'll pick the few best-looking plaques of each of my two phages and plate the phage they contain in new lawns.  From those new lawns I'll again pick the best-looking plaques, and plate their phage in new lawns.  Etc.  Maybe I'll introduce a bit of UV mutagenesis along the way.

The first step will be to make fresh lysates of these phages.  The lawns I made before are too old, so I'll grow up some cells for lawns today and tomorrow I'll retiter the lysates.  On Friday I can pick plaques from these lawns and make plate lysates.  If there's a plate with near-confluent plaques I can use it directly to make a plate lysate.  (10^7 or 10^8 pfu/ml, and I have maybe 5 µl so at best I can get plates with 5 x 10^4 or 5 x 10^5 plaques.  The latter might be enough to get a good lysate.  There are small volumes (50-100 µl?) of the original lysates in the lab upstairs, so maybe I'll sue these.  Or Maybe I should save these until I've improved the plaquing conditions.

thin lawns, feeble or absent phage

My phage titering gave disappointing results.  Three of the five lysates gave no plaques at all, and the other two gave small indistinct plaques that couldn't be accurately counted or characterized.

I took some photos of the plaques I did see.  The top photo is a section of one lawn, with several thousand tiny indistinct plaques.  (The blurry markings are the label on the bottom of the plate.)  The second photo is a closeup  of an area on another lawn where I had spotted more-dilute phage, taken with my iPhone's Olloclip zoom lens.  A few tiny plaques are visible, maybe 3, maybe 5.

For comparison, here's what nice plaques look like.  These are plaques of the E. coli phage lambda (source)

I won't be able to use these R. capsulatus phage for my GTA-vaccine experiments unless I can get better plaques.  I'll need to know whether the phage makes turbid or clear plaques, and I'll need to be able to count it accurately.

I can try using another strain as the host.  These lawns were made with a culture of strain YW1, the strain that these phage were originally isolated on.  I have several other strains, though I don't know if they are closely related.

I can also try changing the plating conditions.  I followed the protocol that I obtained from people who have worked with these phage, but perhaps I could grow the cells to a different density, or incubate the plates at a different temperature.   I'll ask the experts for advice.

Titering my lysates

Planning today's work:

Titering the phage lysates should be a no-brainer, but it's been a long time since I worked with phage so I'd better think things through before I do it.

I have 15 µl of each of 5 phage stocks ('lysates').  The original titers (plaque-forming units/ml, pfu/ml) are written on the tubes - they range from 6x10^5 pfu/ml to 2x10^11 pfu/ml.  But the lysates are probably quite old (maybe 2 years, maybe more), so their titers may have dropped a lot.

I think I'll do one poured-lawn plate of each, using an amount of lysate that should be about 1000 pfu according to the original titer.  And I'll do a spotted-lawn plate for each phage, using undiluted lysate  and a range of dilutions.

I'll dilute the lysates in the same YPS medium I've grown the cells in.  It has calcium and magnesium added so should be fine for typical phage.

About bacterial lawns and phage plaques

This was going to be a post where I do the planning to titer my new lysates today, but it turned into an explanation of how microbiologists use plaques in lawns of bacteria to study phages.

Wait, what's a 'lawn' and what's a 'plaque'?  A lawn is a thin layer of confluent bacterial growth, usually created by mixing a relatively large number of cells (≥10^6) with liquid agar solution ('top agar' or 'soft agar' and pouring the mixture onto the surface of a nutrient-containing agar plate. The top agar is usually at  0.5-0.75%, about half the concentration used for a normal solid plate.  The cells can't move around in the agar, and they grow to high density using the nutrients that diffuse upward from the bottom layer.

If a few of the initial cells were infected with a phage, the phage they release when they die will infect neighbouring cells and kill them, creating a cell-free zone called a 'plaque'.

Here is a detailed drawing of what's happening as a plaque forms:

Initially there's just one infected cell, and sparse uninfected cells in its neighbourhood.  When this cell lyses, the phages it releases can readily diffuse through the agar and infect nearby cells.  While this is happening, the uninfected cells are growing and dividing.  When the newly infected cells lyse, the phage they release add to the local population and infect more cells.  The phage continue to diffuse away, but soon the neighbouring cells become so dense that they stop growing and the phage can no longer replicate in them.  The cells are too big to diffuse through the agar like the phage, so lysis leaves a circular cell-free space called a plaque.  Typical plaques are 1-2 mm across so easy to see with the naked eye.

By counting the number of plaques that form in a lawn of bacteria, we know how many infectious phages were present in the mixture we poured on the plate.  This is the standard way to measure the number of phages (well, the number of 'plaque forming units', PFUs) in a preparation of phage (a 'lysate').

Regular poured-lawn method:  Cells and diluted phage are incubated together in a small volume of liquid (broth or phage-dilution solution) for long enough that most of the phage have attached to the cells.  Then hot liquid top agar is added to the tube and the contents are quickly mixed and poured onto an agar plate of whatever medium best supports lawn growth and plaque formation.  (Quickly so the mixture cools before the cells are damaged.)  The top agar quickly sets, and the plate is incubated overnight at an appropriate temperature for bacterial growth and phage plaque formation.

Spot-titer method:  Cells are quickly mixed with hot top agar (no phage) and the mixture is poured onto agar plates and left to set.  Sometimes the plates can be prepared days in advance, if the cells are happy sitting in the fridge. 10 µl dilutions of phage are then spotted onto the surface and the plates are incubated overnight as before.  If you're gentle you can even streak a drop of lysate across the lawn as you would streak cells on a normal plate, allowing you to grow well-isolated plaques without the nuisance of diluting your lysate.

Other ways we can use lawns and plaques:

Isolating phage from a single plaque:  Often you want to start an experiment with a genetically pure phage lysate that you grew up from a single plaque.  If plaques are well-separated (remember that the phage continue to diffuse out after the plaque forms and the lawn stops growing) you can use a Pasteur pipette to punch out the plaque away from the surrounding agar.  If this plaque is put into a small volume of phage-dilution solution, the many thousands of phages it contains will diffuse out over a few hours (or less) and the phage-containing liquid can be used in your experiment, or to prepare a new lysate whose phage all derive from the one that originated the plaque.

Plate lysates:  Lysates can be prepared in broth, by adding phage to growing cells, watching for the time when the culture clears because most of the cells have lysed, and pelleting out the cell debris.  This is a bit fussy to do, since clearing depends on having the right proportions of phage and cells.  A simpler method is to prepare a 'plate lysate', as follows.  Mix the liquid from a picked plaque or a small amount of a lysate with cells and top agar, and pour a lawn.  You want enough phage that the resulting plaques will be 'confluent' - will overlap just enough that very little intact lawn remains.  Once the plaques have formed, overlay the top agar with 5 ml of phage-dilution solution and leave for a few hours or overnight.  Half of the phage will diffuse into the liquid, and in the morning you just have to collect the liquid and add a few drops of chloroform to kill any cells.  These lysates usually have very high titers, because the cells in a lawn can grow to much higher density than those in a liquid culture.

Phage-resistant colonies:  
Sometimes, the area around an initially infected cell includes a cell that is genetically resistant to the phage due to a new mutation that blocks phage attachment or reproduction.  Such as cell (green in the diagram below) will be able to grow within the  area of spreading phage, and its descendants will form a visible colony within the plaque.

Turbid plaques:  One other phenomenon deserves mention, and that's the 'turbid' plaques formed when a 'temperate' phage infects a lawn.  Temperate phages are those that have a mechanism to enter a dormant state in host cells, where the phage genome is passively replicated by the cellular machinery, usually because it is integrated into the cell's chromosome.  Cells with such dormant phages ('lysogens', orange in the diagram below) are resistant to infection by external phages.  When a temperate phage forms a plaque, most infected cells lyse and produce infectious phage, but some form lysogens that grow and divide within the plaque.  Usually many such cells form causing the center of the plaque to appear cloudy ('turbid') rather than having visible colonies.

Questions about CRISPR-mediated phage immunity

Thursday's post described the hypothesis that bacteria might use gene transfer agent particles to inoculate other cells in the population with fragments of phage DNA, and outlined an experiment to test this.  Now I'm realizing that I need to know a lot more about the kind of immunity I should expect to see if this GTA-as-vaccine hypothesis is correct.

Simplistic outline of the experiment:
  1. Infect GTA-producer strain of R. capsulatus with phage under conditions where the infection is inefficient and few cells lyse.
  2. Remove cells and debris from the culture, to get a supernatant that will contain GTA particles and (unavoidably) some phage particles.
  3. Expose a new culture to the supernatant so cells obtain DNA from the GTA particles, again under conditions where successful phage infection will be minimized.
  4. Wash the surviving cells to remove phage (as much as possible).  Allow time for CRISPR formation if needed.
  5. Expose the cells to a titer of phage suitable for selecting resistant cells.  As a control, also expose cells not treated with GTA.  
  6. Plate to isolate colonies from surviving cells.
  7. Test the survivors for phage resistance.
  8. Compare the frequency of resistance in treated and control cultures.
  9. Test resistant colonies for CRISPR changes.

Things I should find out before I do the experiment:

1.  How efficiently do introduced DNA fragments give rise to CRISPR spacers?  If this efficiency is too low relative to the background rate of phage resistance, I won't be able to detect an effect.  This paper (Hynes et al. 2014, thanks to @AprilPawluk for pointing me to it) might let me estimate the  efficiency.  They exposed cells to a mixture of infectious and damaged phage (damaged by a restriction enzyme in the cell or by prior UV irradiation) at a multiplicity of infection (moi) of 0.1-0.2, and then examined the resulting confluently lysed lawns for phage-resistant colonies.  Unfortunately they only report relative changes in frequency of resistant cells (maxima 16-fold and 6 fold for restriction and irradiation respectively), but in their Methods they mention that the highest frequencies of resistance they observed were about 10^-6.  I don't know if this is for naive cells or pre-exposed cells, but even if it's for naive cells, the max frequency of CRISPR resistance I might expect is only about 10^-5.  This would not pose a detection problem, but it would limit the population-level benefits of the proposed vaccine system.

2.  What fraction of the survivors of a phage infection are genetically resistant, and what fraction of phage resistance arises by non-CRISPR mechanisms?  If most survivors are just lucky, then it might be a lot of work to identify the genetically resistant ones.  In the Hynes et al. experiments, all of the colonies were genetically resistant, and all had new CRISPR spacers.  However this might be quite different for different phages.  If most resistant cells have altered phage receptors rather than phage-specific CRISPR spacers, the effect of GTA-mediated CRISR resistance will be hard to detect.

3.  How quickly does CRISPR-mediated phage resistance arise after exposure to phage DNA?  I don't know.  Cells in the Hynes experiment might have had one or two lytic-cycle durations between being infected by the damaged phage and being infected by an infectious phage.

4.  What fraction of phage infections are abortive and thus could lead to CRISPR immunity to subsequent infection?  Inspired by the Hynes experiment, I can increase abortive infections by UV-irradiating the phage lysate.  (I know how to do this well from previous work.)

5.  How efficiently do phage spacers prevent phage infection?  April Pawluk (via Twitter) says they reduce infection by several orders of magnitude. In the Hynes work acquisition of a phage-derived CRISPR spacer enabled cells to form a colony in a sea of phage.

OK.  I have lysates of five sequenced R. capsulatus phages (from Dave Bollivar via Tom Beatty), and I have the R. capsulatus strain these phages were isolated on, as well as GTA-producing and recipient strains. Time to get to work!

Why GTA genes can't be maintained by 'selfish' transmission

Below is the line of reasoning showing that the genes responsible for producing GTA particles cannot maintain themselves or spread into new populations by GTA-mediated transfer of themselves into new cells.  I initially worked this out with a rigorous set of mathematical equations, but then realized that the problem was so glaringly obvious that math isn't needed.

The main GTA gene cluster is too big to fit inside a single GTA particle, so GTA particles can't transmit DNA that converts a GTA- cell into a GTA+ cell.  Some genes outside the main cluster are also required for GTA production.

But GTA particles can (and do) contain one or more individual GTA genes.  If a fragment containing a particular GTA gene is injected into a formerly-GTA+ cell that is now GTA- because it has a mutated version of this gene, the resulting recombination can restore the cell's original GTA+ genotype.

But these transfer events would not allow GTA+ cells to invade a GTA- population, or to maintain themselves in the face of loss of GTA function by mutation.  That's true for all known GTA systems, even in the simplest (imaginary) case where production of GTA particles requires only a single gene that could easily fit into a GTA particle, as illustrated below.  

Why?  Three factors together require that production of GTA particles reduces the total number of GTA+ cells in the population:

Problem 1:  GTA particles can only be released to the environment if the GTA+ producer cell lyses.  So each production event removes one GTA+ cell from the population.

Problem 2:  The GTA genes in the producer cell are not over-replicated as a phage genome would be, so each production event can produce at most one G+ particle (containing the GTA gene or cluster).  

If all steps occurred with 100% efficiency, problems 1 and 2 would allow, at best, replacement of the lost GTA+ cell with a new one created by GTA-mediated recombination.  But this would not maintain the numbers of GTA+ cells in the face of occasional loss of GTA genes by mutation or deletion.  Nor would it allow GTA+ cells to invade a GTA- population.

Problem 3:   Production of GTA particle production, transmission of their DNA to recipient cells, and recombination with the recipient genome are all likely to be at least moderately inefficient.  Here's a partial list of expected inefficiencies:
  1. Burst size:  Actual burst sizes are unknown, but packaging all the DNA in a R.capsulatus. genome would need 841 particles, which is much larger than typical burst sizes for DNA phages.  Capsid proteins may be limiting, since they would be produced from single-copy GTA genes rather than replicated phage genomes.
  2. Dispersion:  The GTA particles will disperse in the environment, and many will probably not find cells to attach to.
  3. Stability:  Lab preps of GTA particles are unstable in non-optimal storage conditions, so many particles will likely fall apart.
  4. Recombination efficiency:  Only one DNA strand enters the cytoplasm, and some DNA degradation is likely.  The highest observed transduction frequency is only ~4^-4, (theor. max: 1.2^-3) so recombination efficiency is probably only ~0.3.  Recombining in a novel gene will be less efficient than simple strand replacement
  5. Self-conversion:  Some G+ particles may attach to cells that are already GTA+.

Might GTA be a vaccination system for infecting phages?

My work at Dartmouth (to be described in upcoming posts) showed conclusively that genes encoding Gene Transfer Agents (such as the GTA system of Rhodobacter capsulatus) cannot be maintained by 'selfish' transfer of either whole GTA gene clusters or single GTA genes into GA- recipients.  Neither can the GTA genes be maintained by general recombination benefits that can arise when fragments of chromosomal DNA are transferred into new cells.  So, although 'gene transfer agent' does accurately describe one activity of these genes, it cannot be the activity for which they are selected.

The main obstacle to the maintenance of GTA genes, which applies to all the benefits is that any GTA+ cell that actively produces GTA particles cells must die, since cell lysis is needed to release their particles into the environment.  Another obstacle, applying to selfish transfer, is that GTA genes are not over-replicated during GTA production (and are not preferentially packaged), so each cell death can produce only one GTA+ particle. 

I presented these results at the Analytical Genetics conference last week, and asked the other participants if they could think of alternative benefits of producing GTA particles.  Sanna Koskiniemi from Uppsala University made the very interesting suggestion that GTA particles could serve as a syringe, packaging DNA fragments from a phage that's infecting the producer cell and transferring these fragments into other as-yet-uninfected cells, where they could trigger development of CRISPR immunity.

I love this idea and want to test it.  It doesn't overcome the cell-death obstacle, but it does overcome the selfish-transfer obstacle since a single producer cell could produce many particles of phage DNA from a single phage genome, and more if the phage genome is replicated.

One way to see if this could provide sufficient benefits to maintain the GTA genes is by simulation modeling like that I used to examine the recombination benefits.  This could clairfy the important factors that would need to be examined.

Here I want to start considering experimental tests of this hypothesis.

The ideal test would be to infect the GTA-producing strain with a phage, preferably under low-growth conditions where phage infections are often abortive.  (Luckily R. capsulatus produces most of its GTA under such conditions.)  Then some recipient cultures would be exposed to the GTA-containing culture medium (and some not, as controls), and then all exposed to a lysate of the phage.

"But wait!", you say.  "Won't the GTA-containing culture medium also contain some phage?"  Yes, probably.  I don't think there's any way to inactivate the phage particles without also inactivating the GTA particles, or vice versa.  We might be able to come up with either perfectly-abortive infection conditions (where infected cells don't produce any phage), or a cellular mutation that prevents phage production.  If not, we might have to combine the GTA-exposure and phage-infection steps.

"And won't any phage lysate also contain some GTA particles?"  Yes, probably.  But we could use a GTA- mutant as the host for lysate production.  Not the mutant that can't lyse, but the one with the main GTA gene cluster completely deleted.

What resources are available for this project?  First I checked with my GTA colleagues, who confirm that R. capsulatus does have a CRISPR-Cas9 system.  Then I asked if there were any well-characterized phage systems able to infect R. capsulatus.  Until quite recently the answer would have been 'No', but a recent paper reported the isolation and sequences of 4 R. capsulatus phages.  A Mu-like phage of R. capsulatus has also been characterized, but it did not form plaques on SB1003.

The report about the 4 new phages used a different host strain (YW1-derived, not SB1003), so the first thing I'll need to do is check whether they form plaques on SB1003.  Then I'll need to play around with infection and plating conditions...  My idea of fun!

Model of GTA evolution by infectious transfer

Here's the description of my model addressing Explanation 1 for GTA persistence.  For now I've just pasted in the text of a Word file I prepared about 10 days ago.

A constant-population-size model of large-head GTA transmission
(Based on Xin Chen’s model, but with stepwise generations and without logistic growth.)

The population:
1.     Population size is constant.  Loss of GTA+ cells due to lysis during GTA production is made up by growth of all cells after the transduction step.
2.     Dense, well-mixed culture in liquid medium (so cells frequently encounter GTA particles)
GTA production:
3.     GTA particles come in two sizes.  Small particles contain 4 kb DNA fragments.  The hypothetical large particles contain fragments that must be at least 14 kb (the size of the GTA gene cluster) but could be as big as 50 kb. 
4.     The number of GTA particles a cell produces does not depend on the proportion of small and large particles.
5.     DNA packaging by GTA is random; all parts of the cell’s genome are equally represented.  But in this model we only consider the particles containing the full-length GTA cluster.
6.     This is the killer:  If the cell’s chromosome is 5 MB and the large-particle capacity is 15 kb, only 2x10-4 of large particles will contain complete GTA gene clusters (will be G+ particles).  If we change the large-particle capacity to 20 kb, then about 1x10-3 of large particles will contain a complete cluster.  A 50 kb capacity and a 3 MB chromosome would probably get it up to about 10-2.  (And this ignores the recombination machinery’s need for homologous DNA flanking the GTA cluster to promote recombination.)
7.     GTA- cells completely lack the main GTA gene cluster.  They can only be converted to GTA+ by homologous recombination with GTA-containing DNA from G+ particles.
8.     GTA particles cannot tell the difference between GTA+ and GTA- recipients.  Particles capable of transducing GTA- cells to GTA+ can also ‘transduce’ GTA+ cells to GTA+.
9.     All GTA particles produced in one cycle are taken up by and transduce cells in that cycle.  (The efficiency of infection and recombination is 1.) 
10.  The model ignores large and small GTA particles that don’t transduce GTA+.
11.  Each cell takes up only one G+ particle (or none).  This is reasonable, since the number of G+ particles is always going to be much smaller than the number of cells.

F    Initial frequency of GTA+ cells (we want to consider a wide range)
c    Fraction of GTA+ cells producing GTA particles (and consequently lysing).  (In wildtype lab cultures this is <3 o:p="">
b    Number of GTA particles produced by each burst.  Default value is 100.  (We have no actual measurements.)
µ    Fraction of GTA particles that are large.  (We expect this fraction to be small, since large particles have not been observed.)
T    Fraction of large GTA particles that are G+ particles (able to transduce GTA).  (This is limited by genome size, GTA gene cluster size, and the DNA capacity of these hypothetical particles.  Plausible values are between 10-2 and 10-4.)
G   µ * T Fraction of GTA particles that contain complete GTA genes.

What happens in one generation:
GTA production and cell lysis:
N   Proportion of GTA particles to cells remaining in the medium after GTA+ cells have burst. 
      = (Fcb)/(1 – Fc)  (Note: Fcb is the GTA production per original cell.  1 – Fc normalizes this to the number of cells remaining after lysis.)
N+  Proportion of GTA particles, per remaining cell, that carry the complete GTA gene cluster (are ‘G+’ particles able to transduce the GTA-production genotype to GTA- cells). 
= NµT   =  NG
Fraction of surviving GTA+ cells per original cell (will be normalized to remaining cells later): = F(1 – c)
Fraction of GTA- cells transduced to GTA+: N+(1 – F).  {Note: the 1 – F corrects for the G+ particles that attach to and ‘transduce’ GTA+ cells.) 
Fraction of GTA+ cells (per original cell) after transduction:  F(1 – c) + N+(1 – F).  (Note: F(1 – c) removes cells killed by lysis, N+(1 – F) adds cells gained by transduction.)
Fraction of GTA- cells (per original cell) remaining after transduction:  (1 – F) – N+(1 – F).  (Note: 1 – F is the original fraction of GTA- cells, N+(1 – F) removes cells lost by transduction to GTA+.)
Cell growth:
Now we normalize the cell numbers to ‘per remaining cell’:
Total fraction of cells remaining after GTA production and transduction:
            1 – (Fc)  (Note: To normalize, divide the above cell fractions by this value.)
Fraction of GTA+ cells after one complete cycle:
F’ = F(1 – c) + N+(1 – F) / 1 – Fc

How to evaluate the change in the proportion of GTA+ cells?
We can expand N+ and pull out the F, then look at the before/after ratio:
F’   = F * (1 – c) + c * b * F * µ * T * (1 – F) / 1 – (F * c)
      = F * ((1 – c) + C * b * µ * T * (1 – F) / 1 – (F * c)

F’ / F = (1 – c) + c * b * µ * T * (1 – F) / 1 – (F * c)

When the value of this expression is greater than 1, GTA+ is increasing; when it is less than 1, GTA+ is decreasing.
For simplicity, below I combine b, µ & T as the compound variable W.

What happens if we vary F, holding everything else constant?
Increase of GTA+ depends only on W.  If W is >1, GTA+ increases.  If W is <1 decreases.="" gta="" o:p="">
The rate of change is very slow when F is close to 1 (when almost all cells are GTA+), and fast when F is close to 0 (when almost all cells are GTA-).
What happens if we vary c, holding everything else constant?
C affects how fast change happens, but not its direction.  If W>1, GTA+ still spreads; if W<1 decreases="" gta="" o:p="" still="">
What happens if we vary W, holding everything else constant?
If W<1 always="" be="" denominator.="" numerator="" o:p="" smaller="" than="" the="" will="">
If W>1, the numerator will always be smaller than the denominator.
In both cases., all the other parameters cancel out.  This confirms that the direction of selection o GTA+ depends only on whether W is higher or lower than 1.
Would the result change if the population were growing?
I don’t think so, since GTA+ and GTA- cells grow at the same rate.

Since plausible values of W are all much lower than 1, I conclude that GTA+ cells cannot increase by GTA-mediated transduction of GTA- cells to GTA+.

GTA could spread by transduction if it did preferentially package the GTA gene cluster into its particles.  Of course, then it would be a phage.
How the model’s assumptions affect this outcome:
Basically, all the assumptions are either neutral or increase the chance that GTA+ will spread. Making the simulation more realistic would just make things worse for GTA+, not better.
The population:
1.  Population size is constant.  Loss of GTA+ cells due to lysis during GTA production is made up by growth of all cells after the transduction step.
I don’t think adding growth would affect the outcome.
2.  Dense, well-mixed culture in liquid medium (so cells frequently encounter GTA particles).
If the culture were more dilute or poorly mixed, some GTA particles would not find new cells to attach to.  This would reduce the amount of transduction (effectively reducing W).
GTA production:
3.  GTA particles come in two sizes.  Small particles contain 4 kb DNA fragments.  The hypothetical large particles contain fragments that must be at least 14 kb (the size of the GTA gene cluster) but could be as big as 50 kb. 
This is the central assumption of the model.  The size of the small particles is known.  The hypothesized large particles could be as small as 15 kb (allows a bit of homologous sequence on each side of the cluster to promote recombination).  Phage capsids can in principle be very large, but it’s parsimonious to assume a modest size.
4.  The number of GTA particles a cell produces does not depend on the proportion of small and large particles.
Large capsids will require more capsid protein molecules.
5.  DNA packaging by GTA is random; all parts of the cell’s genome are equally represented.  But in this model we only consider the particles containing the full-length GTA cluster.
Experimental results show slightly less packaging of GTA sequences.  If this applies to the hypothetical large particles it would reduce production of G+ particles.  If particles preferentially package GTA, GTA would be a phage.
6.  This is the killer:  If the cell’s chromosome is 5 MB and the large-particle capacity is 15 kb, only 2x10-4 of large particles will contain complete GTA gene clusters (will be G+ particles).  If we change the large-particle capacity to 20 kb, then about 1x10-3 of large particles will contain a complete cluster.  A 50 kb capacity and a 3 MB chromosome would probably get it up to about 10-2.  (And this ignores the recombination machinery’s need for homologous DNA flanking the GTA cluster to promote recombination.)
See point 3 above.
7.  GTA- cells completely lack the main GTA gene cluster.  They can only be converted to GTA+ by G+ particles.
Transduction depends on homologous recombination.  Small GTA particles can transduce functional alleles of individual GTA genes, replacing versions that became mutated or even deleted in an ancestor of the recipient cell.  But they cannot introduce GTA genes into cells that completely lack the GTA cluster, because there will be no homologous sequences to recombine with.
8.  GTA particles cannot tell the difference between GTA+ and GTA- recipients.  Particles capable of transducing GTA- cells to GTA+ can also ‘transduce’ GTA+ cells to GTA+.
I think some phages and conjugative plasmids may be able to detect whether potential hosts/recipients already have the element, but we have no evidence that transduction frequencies differ between GTA+ and GTA- recipients.  Wall et al (1975) surveyed 33 strains and found wide variation in both GA production and transduction, but no correlation between these abilities.
9.  All GTA particles produced in one cycle are taken up by and transduce cells in that cycle.  (The efficiency of infection and recombination is 1.) 
This is unlikely to be true, but assuming this increases the chance that each G+ particle successfully transduces a GTA- cell to GTA+.
If we were to relax this assumption the model would need to include an explicit uptake process and to specify what happens to particles that are not taken up.
10. The model ignores large and small GTA particles that don’t transduce GTA+. 
This should be OK, since these should not interfere with transduction by G+ particles, especially because their total number per cell will be small. Removing this assumption would make GTA + spread less likely.
11. Each cell takes up only one G+ particle (or none). 
This is a reasonable assumption, since the number of G+ particles is always going to be much smaller than the number of cells.  If the number of G+ particles were high, sometimes two G+ particles might inject their DNAs into the same s=cell, which would reduce the efficiency of transduction.