Reviewing Data and Revising Hypotheses

Hi blog, it’s been a while. I recently gathered evidence (via restriction digest) that a pair of double mutants I’ve been using (sxy-1 Δhfq and murE749 Δhfq) were swapped at some point. Since then, I’ve been carefully evaluating all the data I’ve collected and diving into the literature to try to make sense of everything. My goal was to come up with a concrete hypothesis that explains what I have seen and I think I have one. But first, we will look at the data:

If you get lost at any point, I have provided a little bit of background at the bottom of this post.

(order of the bars in the graphs is KW20, sxy-1, murE749 for any colour-blind readers out there)

First, we’ll look at murE749. We really don’t know much about why this mutant is so competent, but previous work in our lab has shown that the murE mutant expresses the same amount of Sxy protein as KW20, but has the Sxy-dependent competence genes turned on when they are typically off. (http://www.zoology.ubc.ca/~redfield/PDF/Ma&Redfield%20murE.pdf) The murE mutant also shows no competence when sxy is removed. This suggests that the effect of the murE mutation takes place somewhere downstream of Sxy in the competence-inducing pathway and its effect is dependent on the presence of Sxy. Looking at my data, it seems that removing hfq makes no significant difference to how competence murE749 is (except for, perhaps, the late-log + cAMP condition). I understand this to mean that whatever the murE mutation does, its effect supersedes the negative effect of removing hfq.

Turning to sxy-1, we see something similar. There may be a tiny reduction in competence under most conditions, but for the most part it does not seem very strong or even statistically significant. Competence is strongly induced when it is not normally in KW20 (see log phase) presumably because of higher translational efficiency of basal levels of sxy mRNA.

Finally, the most interesting results are with the single Δhfq mutant. We see a 10 fold reduction in cells lacking hfq when grown in MIV. At low density, cells with and without hfq show similar levels of competence. However, as the culture gets dense we fail to see any induction of competence in Δhfq while KW0 becomes about 50 fold more competent. The story is different when exogenous cyclic AMP is added though. The data suggests that exogenous cAMP temporarily rescues competence (log phase + cAMP).

To illustrate this further, here’s time course data of KW20 and Δhfq with and without cAMP:

This data illustrates a few things:

• Without cAMP, KW20 can become 100 fold more competent than Δhfq
• KW20 becomes more competent as cells get dense, see little to no increase for Δhfq
• Competence in both strains drops as time continues
• With exogenous cAMP, both strains become almost equally competent, perhaps KW20 slightly more than Δhfq
• Competence drops off back to no cAMP levels at about OD = 0.1 (also seen in previous set of graphs)
• Knowing all this, what conclusions can be drawn? I think it’s very likely that Hfq has a role in cAMP regulation.  

If we are to assume that Hfq in necessary to fully induce the production of intracellular cAMP, the data begins to make sense. As cells become dense and the nutrients in the medium in which they are growing become depleted, competence increases presumably due to activation of the pathway that induces cAMP production. Perhaps by removing hfq, cAMP levels fail to increase to such an environmental influence. Giving cells exogenous cAMP removes the influence of intracellular regulation, and under this condition, we see competence rescued (at least for a period of time) in the hfq knockout.

This makes me think that the influence Hfq has on competence happens before the transcription of sxy. Hfq’s function could be to:

• increase the sensitivity of the phosphotransferase system
• increase translation of adenylate cyclase (increase cAMP levels)
• reduce translation of cAMP phosphodiesterase (prevent depletion of cAMP)
• increase translation of crp (increase response to cAMP)

I don’t think that CRP is regulated by Hfq as it is in Y.pestis. If Hfq regulated CRP at the transcriptional level, it would be odd that adding exogenous cAMP would somehow remove that layer of regulation.

I would be surprised if translation of protein in the phosphotransferase system was regulated by Hfq merely because it would odd to affect sugar uptake to indirectly change cAMP levels in the cell. There may be something here I’m missing but my initial thoughts are that this should not be my focus.

So in terms of a hypothesis, I’m favouring adenylate cyclase. cAMP phosphodiesterase is another option, but I think it makes less sense. To use an analogy: if you want a room dark, it makes more sense to prevent a light switch from being turned on rather than standing there waiting to turn it off every time it comes on.

So here’s my current hypothesis:

Removing hfq causes a competence defect by reducing adenylate cyclase translation and therefore reducing intracellular cAMP levels, preventing the induction of competence.

I think this hypothesis aligns with what is seen between Δhfq/KW20. The murE mutant appears to be consistently hypercompetent whether or not cAMP is present, suggesting that it is independent of adenylate cyclase activity and it makes sense that little to no difference is seen when hfq is removed. As for sxy-1, even basal levels of sxy expression are enough to induce competence, downplaying the role of cAMP in competence regulation. Perhaps that’s why the difference between the mutant and double mutant is almost trivial.

I did want to talk about future experiments, but this post is already extremely long, so I’ll do that in a separate one.

Some Background:

About the strains

KW20 is the Rd Haemophilus influenzae strain. Δhfq is KW20 with the hfq gene removed with a spc resistance gene added. sxy-1 is the Rd strain with a point mutation that increases translation of a gene necessary for transcribing the competence regulon. This mutation increases competence by removing an inhibitory structure in the sxy mRNA and inducing higher rates of translation of the Sxy protein. murE749 has a point mutation in an enzyme required for peptidoglycan biosynthesis that increases competence by some unknown mechanism. Double mutants carry both mutations. Error bars indicate that replications have been done, typically 2 or 3 times.

About competence and cAMP

When H. influenzae is starved, the phosphotransferase system detects a lack of preferred sugar and activated adenylate cyclase (cyaA). Adenylate cyclase then catalyzes the production of intracellular cAMP. This cAMP binds to C-reactive protein (CRP), this complex then induced the transcription of CRP-dependent genes.  This set of genes involves sxy, which when induced activates the competence genes.

The mystery solved?

Well, I think I get it now. I was getting an odd phenotype for the Δhfq sxy-1 double mutant that suggested that this mutant was more competent than the sxy-1 single mutant. This is weird since knocking out Hfq in all other conditions tested so far has a negative effect. I tested the other hypercompetent sxy point mutations in a Δhfq background and again, I only see a negative effect:

And of course...

This has been eating at me for a long time now. Last lab meeting, I think it was Rosie who asked whether or not the Δhfq sxy-1 strain was accidently mixed-up with the Δhfq murE749 strain. Thinking about it more, I realized that when Rosie had originally preformed the initial characterization of the Δhfq mutants, her lab notes had swapped the two strains in some places. Looking at the data, it also seems reasonable that what I’ve been calling Δhfq murE749 is actually Δhfq sxy-1:

I’ll have to experimentally verify this by using a restriction endonuclease (Mnl I) that has an additional cleavage site in the murE749 mutant. It’s nice to know that this will probably answer a lot of questions that I have, but I am now left with no leads to identify where the Hfq effect on competence occurs. Science is a harsh mistress.

I don't even know what to think about this.

I've been recently thinking hard about how hfq influences competence and I'm stumped. Here are the facts:

Removing hfq in KW20
- Competence is lowered 10 fold when cells are in MIV
- Competence fails to increase as cells become dense (transformation frequency remains <5E-6)
- Adding cAMP temporarily increases competence back to expected wild type + cAMP levels but drops off again quickly

Removing hfq in sxy-1
- Competence in MIV is pretty much the same between the sxy-1 mutant and the double mutant
- Competence reaches MIV levels when growing in log phase (>10x higher than single mutant sxy-1)
- Double mutant in general is more competent than single mutant
- When adding cAMP cells appear to become maximally competent with about 1 in 100 cells transformed

Apparently adding a mutation that lowers competence in wild type to a hypercompetent mutant with a single point mutation increases competence even more so? I don't get it. As if it didn't get confusing enough....

Removing hfq in sxy-2
- Competence decreases about 5 fold for the double mutant in MIV
- Competence is a little lower for the double mutant in log phase (~1E-4) and fails to increase as the cells become more dense
- Not sure about the effect of cAMP yet.

The mutations in sxy-1 and sxy-2 are literally under five bases apart and when added to the wild type background, have a similar phenotype. Yet when we remove hfq from the sxy-1 mutant we see increased competence under almost every condition and when we remove hfq from the sxy-2 mutant, we see decreased competence. I think this is strong evidence that hfq may directly interact with the sxy mRNA in some capacity: the question is how?

Haemophilus sRNA?

I was looking at the old RNA seq data, specifically regions between genes, and I found an enriched region between the gene for CRP and the upstream gene, it looks like a potential small RNA. Here:

It seems like it's actually differentially expressed over time (note, bottom graph is log scale) which is why I chose to closely look at this region in particular. The coverage is strikingly consistent at it's peak. I put the sequence into Rfam and I got a single result for the 3' end: C4 antisense RNA. It appears to target the ant (adenine nucleotide translocator) gene and originates from a phage. As far as I know, Haemophilus influenzae does not have an ant gene. It does appear to mostly follow the structural template for this RNA as seen below:

I hope to find other possible sRNA using a similar technique. Stay tuned.

RNA seq Analysis

I took a look at some RNA seq data that was previously acquired by a post-doc in our lab and I've been doing various things to just practice analyzing large datasets. Here's some interesting things I've done:

(note: K = KW20, U = HI_0659-,  T = HI_0660-,  S = sxy-)

Coverage of sxy mRNA

Nucleotide-resolution of the sxy mRNA reads. There appears to be less expression in the HI_0659 knockout and possibly more in the HI_0660 knockout. The two KW20 replicates appear to differ but I think it is a timing issue since 10 min and 30 min lines in the B replicate appear to be averaged versions of the 10 min and 30 min lines in the A replicate.

 

 Number of fragments that start/end at a particular base (KW20)

The common drop around base #255 (especially at 30 min) interested me for quite a bit as a possible cleavage site but it could also be due to specific secondary structural elements that encourage strand breakage during fragmentation. I plan on creating graphs like these for the rest of the genome so I'll see if this is a common thing.

Timecourse for all genes

I wrote an R script that creates a timecourse-type graph for all the genes under every condition. Here's what they look like:

The competence genes actually stand out because the strains really differ at the 30 min mark:

I also tried normalizing the score a couple different ways. First I normalized by the sum of the fragment counts for the whole genome for each condition, second I used the value at the 75% quartile for each condition. Here's a comparison (chosen at random):

 

1. Normalized by sum

2. Normalized by quartile

I think I prefer the second method.

Finally, one interesting thing I noticed was that the trp operon appeared to have similar characteristics to the competence proteins:

 One trpR regulated operon

  Another trpR regulated operon

trpR

I'm not sure if there is a known connection between the trp operon and competence regulation? There appears to be according to these graphs (and others). Perhaps it's something to pursue.

Coming up next...

I just want to clarify where I'm currently at and what paths I can take from here.

So I've completed my big transformation experiment as described in previous posts. Here's the resulting data:

The first two graphs show the difference in transformation frequency between strains and under different conditions. The second pair of graphs shows the ratio between transformation frequency in hfq+ and hfq- strains.

A few things stand out and I would like to pursue them.

1) The sxy-1 hfq- double mutant apparently has a higher transformation frequency than the single sxy-1 mutant. This is striking. I would like to see if this is consistent across other sxy point mutations. I'm in the process of making a sxy-2 (hypercompetent) hfq- mutant to see if this is consistent. I would also like to construct a sxy-7 (decreased competence) hfq- mutant and see the effects of strengthening the mRNA structure because I'm starting to think there may be some regulation happening at the RNA level. The problem with sxy-7 is that it decreases the natural transformability of the cell, so it's much harder to get the hfq knockout into it's genome. I'll see if I can find some way.

2) For some reason, in the simple hfq- strain, there appears to be a consistent TF across all conditions except MIV and log + cAMP. Something must be going on to cause the TF to fall after adding cAMP as the culture gets dense. I would like to do a time-course experiment for the hfq- mutant and perhaps try adding the cAMP at different times to see what happens.

3) I would also like to try and see where the effect of hfq is happening on the competence pathway. Back in the 90's, someone in the lab created a strain that expresses beta galactosidase in the presence of the CRP-cAMP complex. I would like to see how the expression of this reporter gene changes in the hfq knockout verses an hfq+ strain. I am currently growing the hfq+ version of this strain and I plan on transforming hfq- DNA into it.

That'll give me something to do.

Why Fructose?

Today I learned that only the fructose phosphotransferase system (PTS) is present in the Rd strain of Haemophilus influenzae. This means that H. influenzae cells rely on fructose as their primary source of sugar. This is odd considering glucose is much more abundant in respiratory mucus where we would expect to find H. influenzae (meaning it would make more sense if the cells relied on glucose rather than fructose). When the cell's preferred sugar is not available, the cell turns on genes that allow use of other sugars (and also, the cell turns on it's competence genes and becomes ready to uptake DNA).

I was just thinking about this and I hypothesized that perhaps the cell relies on fructose because it wants to turn on it's competence genes often - thus relying on a scarce sugar would keep the cells competence machinery in production more often. I also thought that perhaps it's actually more effective to use DNA as a source of energy than glucose, so I took to the internet and I tried counting expected ATP output for glucose vs. nucleotides:

Glucose
+2 net ATP from glycolysis
+2 ATP from citric acid cycle
+32 from electron transport chain (let's say)
36 net ATP production from glucose breakdown

To transport glucose using a PTS system, it requires an equivalent of 2 ATP, however one of these reduces the amount of ATP needed in glycolysis by 1 so it costs 1 ATP to transport glucose into the cell.

Nucleotides
I'm going to assume the cost of breaking dsDNA down into deoxyribose and nitrogenous bases is 0 which may be entirely wrong.

Deoxyribose
-1 ATP to split into glyceraldehyde-3 phosphate and acetylaldehyde
+18 ATP since glyceraldehyde-3 phosphate undergoes half of glucose catabolism
+1 ATP to balance the fact that an early energy costing step of glycolysis was skipped
+14 ATP produced from acetyl-coA and NADH produced from acetylaldehyde
32 net ATP production from deoxyribose breakdown

Nitrogenous base
De novo synthesis of a base is 4 ATP so I will say that the base is worth +4 ATP since rapidly dividing cells are always in need of bases.

That makes 36 ATP molecules per nucleotide which means the two break even in terms of energy yield. The real question is how much energy does it take to uptake DNA per base? I don't know if the answer's known. If it's less than 1 (since entire strand are taken up), than it's possible that using DNA would in fact a more effective way to produce energy than sugar.

I don't think that the cells would rely on DNA over sugar for energy, but it was fun to do the accounting. I know there's a lot of DNA in mucus and that any sugar supply would be competed for by the host cells, so there is some reasoning behind this idea, but again, I'm not really convinced.

Also, to get these ATP values, I had to look at biochemical pathways and just count the reactions that required ATP. There may be errors (ie. that fact that I was using mostly human pathways since more information is available) so if you see them, let me know with a comment.