CRISPR

We decided to create a model of the CRISPR system for two main reasons:

• Identifying the parts of the network that could be targeted by our lab team to improve repression efficiency
• To approximate time-series mecA repression data for use in modelling the overall vulnerability of a S. aureus population

The steps we followed are detailed below, but were:

1. Model Formation
2. Parameter Finding
3. Model Formation
4. Parameter Adjustment
5. Sensitivity Analysis

Model Formation

After a literature review we were able to construct the CRISPR interference system network. The targeted single guide RNA (sgRNA) associates with nuclease-deficient Cas9 protein (dCas9) to form a complex that binds with the DNA complementary to the sgRNA target . The bound complex prevents transcription elongation by RNA polymerase, repressing YFP mRNA expression . The chemical network is shown below:

Using standard mass-action kinetics, the network simplifies into the following set of differential equations:

We chose the model kinetics to be largely first-order; this decision was supported by the findings of several recent studies . To simplify the model, we assumed that the formation of the dCas9-sgRNA complex ($b$ in Figure xyz) is in made a quasi-steady-state. That is, we assume that the association/dissociation of dCas9 and sgRNA occurs on a faster timescale than the other reactions in the network (i.e. transcription, translation and the binding of the complex to the DNA), allowing us to assume that the complex is always at steady-state, relative to the other time-dependent species concentrations. This same assumption was made in previous modelling efforts, e.g. .

Under this quasi-steady state assumption, the differential expression for the complex is given by:

Our model then simplifies to:

This is the same assumption made by previous teams.

Modelling Incomplete Repression

A recent study by Bikard et al. found that maximal repression (on the order of 100 fold) was achieved when the promoter was targeted. However, targeting the promoter is not viable in this project since an essential promoter from elsewhere in the genome has been harnessed to produce the fluorescent promoter. Instead, we model the incomplete repression (ranging from 6-fold to 35-fold) observed when the off-promoter regions, specifically on the non-coding strand, are targeted.

There are two possible approaches for modelling the incomplete repression, each reflecting a different physical mechanism that allows leaky YFP expression. In the first mechanism, RNA polymerase is sometimes able to cleave the bound dCas9-sgRNA complex from the DNA. In the second mechanism, the complex binds inefficiently and is sometimes separated from the DNA, permitting transcription to continue.

We assumed that the incomplete repression is accounted for by the first mechanism. This assumption was based on several studies showing radically different repression rates if the complex targets the promoter, preventing transcription initiation, rather than targeting the DNA further downstream and impeding transcription elongation. The differences in the system behavior depending on whether or not RNA polymerase has the opportunity to bind suggest that the “cleavage” mechanism may more closely resemble the chemical reality.

Consequently, we modeled incomplete repression using a leaky expression term proportional to the expected YFP expression when the complex is saturated. The differential equation model was updated with a repression term dependent on the fold reduction FR and the initial concentration of YFP mRNA, Y0:

This equation was derived using two boundary conditions. Before repression, when the concentration of the complex is zero, YFP mRNA is produced at the rate expected from the sarA promoter, α. After repression has reached its steady state, the YFP mRNA production has been reduced by FR fold, to Y0/FR.

Parameter Finding

We turned to the literature to find parameters for our model, given in the Table below. We first looked for parameter values that had been measured in S. aureus. In cases where those could not be found, we next looked for ways to to estimate the parameters using other available data for S. aureus and finally searched for the parameters in other gram-positive bacteria. Aggregating parameters from many experiments across the literature is by nature a somewhat uncertain endeavor; those parameters about which we are very uncertain are marked with asterisks. An explanation for how we arrived at each parameter is given in the table, but details on the more circuitously estimated parameters are given after the table.

The only model parameters without some basis in the literature are the association rates for dCas9 and sgRNA. However, we have made a quasi-steady state assumption for that reaction, which requires that it reach equilibrium on a much faster time scale than the rest of the system. We thus defined a range for the possible values based on the other model parameters

Details on the more roundabout estimations are given below:

Production of dCas9 from dCas9 mRNA

We were unable to find a peptide chain elongation rate for S. aureus, so instead we used the values reported in BioNumber 107869 which gives a range of 0.59-3.17 amino acids per second per ribosome in Streptomyces coelicolor, another gram-positive bacteria. Freiburg's dCas9 part from last year is composed of 1372 amino acids. This translates to a range of 0.0258 to 0.1386 dCas9 molecules per minute per ribosome.

We were unable to find ribosome densities in S. aureus, but found two different estimates for ribsosome density in Bionumbers: 0.22 ribosomes per 100 codons (i.e. per 3 nt coding sequence) and 3.46 ribosomes per 100 codons . Using our assumption of 3 nt:1 amino acid, we then multiply to get the 0.0057-0.4797 range of dCas9 molecules per minute.

Degradation rate of dCas9

We were unable to find any specific data on dCas9 degradation, so instead we used a protein half-life of sarA measured in S. Aureus by Michelik et al. . We chose sarA rather than a protein more chemically similar to dCas9 because data on sarA was readily available and because dCas9 is transcribed using the sarA promoter, which allows us to at least capture sensitivity of the degradation rate to production.

mRNA production from the sarA promoter

We used the time-series data given by Cheung et al. to estimate the rate of production from the sarA P2-P3-P1 promoter in S. aureus. The figure from their paper is reproduced below. After diluting 1:100, the S. aureus strains were serially monitored for OD_650. We used data from the sarA+ strain, as that's more like a wild-type S. aureus strain.

Using the laboratory-conditions doubling time of 24 minutes given in given in , we found that the bacteria would re-enter stationary phase after 2.5 hours; for time-points after 3 hours, the number of number of sarA genes producing fluorescence could be assumed as constant. For this reason, we excluded time-points prior to 3 hours. We then converted from fluorescence units to number of fluorescent molecules using the quantization measurements provided by Wu & Pollard and, using our assumption of a fixed number of active sarA genes, considered the relative change in number of molecules to be representative of the per-promoter rate.

We were interested, however, in the changes of concentration rather than the changes in the raw number of molecules. As the name suggests, Staphylococcus aureus are spherical in shape. Assuming that all S. aureus are spheres, the volume of the cell can be determined. The diameter of a USA300 S. aureus cell was previously measured as 1.1 μ•m resulting in the overall cell volume to be calculated as 5.575•10^-15 L. The number of molecules were thus converted to units of molar concentration in the cell, specifically nanomoles per litre (nM). The exponential fit used to find the rate constant is shown beside the figure from Cheung et al. above.

This resulted in a exponential model a•ebt with a b rate constant of 0.0011 nM/min.

Initial Model Results

Using these estimated parameters, we simulated our differential model of the CRISPR system. It was immediately clear that these parameters, scraped from assorted publications, did not provide an accurate system model once combined. Below are two simulations of the system dynamics using the parameters found in the literature. On the left the system is shown with CRISPR repression active and on the right is a simulation where:

or rather, the system simulated without any repression of YFP Transcription by CRISPR. Both plots use 6 as the expected final fold reduction.

The difference between the two plots is negligible. This unrealistic behavior emerges directly from the parameter values. The mRNA degradation dynamics prevent sgRNA from reaching any significant concentration, so there is never enough dCas9-sgRNA complex to influence YFP expression. However, in both plots the YFP mRNA levels plummet from the high initial concentration because the degradation rate is so much higher than the production rate.

These plots do not fit with the observations of CRISPR interference systems reported in the literature . To ensure that our parameters were at fault, rather than the fact that we were examining CRISPR in S. aureus instead of E. coli, we consulted fluorescence results from the laboratory, which showed that unrepressed YFP continued to fluoresce after several hours.

Parameter Adjustment

Sensitivity Analysis

sRNA

Relevant Biology

Model Formation

Conjugation