Note: this is a search engine friendly version of my lab notebook, please see the pdf version of this document for a more human friendly, printer friendly version.

Chapter 2
Towards a faster, more reliable ChIP protocol

THIS CHAPTER/PROJECT IS NEAR COMPLETION
Check out my blog post "Factorial and response surface optimization of a chromatin immunoprecipitation protocol" for a more in-depth introduction of my goals for this project. Also, if you have questions or comments, please post them on the blog as well.
In Chapter 1, I got a ChIP protocol hammered out well-enough to find a known targets to the transcription factors Lrp, PdhR, and FecI. The protocol is dreadfully slow and tedious however. If I could shorten it and produce at least as good separation between enriched and random then adding more replicates would be less painful and would produce more reliable results.
The hope is that by doing a couple rounds of factorial experiments with 4-8 factors, I'll be able to provide better enrichment in much less time. Hopefully, filter based bead-washing methods or dynal magnetic beads will pan-out, allowing me to remove one of the easiest places to make an error (i.e. sucking your beads out of the tube). After getting the protocol shortened, might be possible to use response surface methods to increase enrichment even further? Would be a cool tech paper if it worked.
The methods I'm using to shorten/improve this protocol are described in the excellent book: Statistics for Experimenters. Box, Hunter, Hunter. If you like history and old books, Fisher developed and wrote quite a bit about the experimental designs I use in this chapter.
Brief Update Thu Dec 13 15:10:17 EST 2007:   Two much improved protocols have resulted from this chapter: one requiring 1.5 days and one requiring 2.5 days. Both of the protocols are 96-well format capable. The protocols can be found on J's Blog in the post: "Optimized ChIP Protocols" (http://blog-di-j.blogspot.com/2007/12/optimized-chip-protocols.html).
Brief Update Wed Aug 23 18:17:53 EDT 2006:   I originally planned to use the more enriched DNA from this protocol for SAPE. Now I hope to be able make a Paired-end-tag library to just sequence it straight away with highly parallel sequencing.
Here are some things I'd like to achieve:

2.1  ChIP improvement round 1: shortening the time

This round will mainly focus on seeing if shortening a few bottleneck steps without worsening performance 3. A few things that might improve the noise enrichment vs true enrichment ratio have been include as well and are 1) altering antibody conc and 2) altering the formaldehyde concentration
Here is the factor list (the 0-state is what I've used in the past):
factor 0-state 1-state
plasmid low copy high copy
formaldehyde conc. 1% 0.1%
quench with glycine yes no
quantification part 1 use qiagen (requires changing volumes) no yes
shearing 4 x 20% x 30 sec 3 x 10% x 20 sec
preclear yes no
antibody conc (per 25 mg DNA) 2 mg 6 mg
antibody incubate time overnite 1hr
bead incubate time 2hr 30min
wash method pellet filter column
final cleanup phenol:chloroform qiagen
bead type agarose dynal
Other than these factors, all other aspects will remain as similar to the protocol in section as possible.

2.2  First round ChIP optimizations

I'm doing a 16 sample factorial design with lrp (the TF I have cloned with the most known targets).
The factors are (low - high):
  1. IPTG (0.01 mM - 1.0 mM )
  2. formaldehyde (0.1% - 1%)
  3. quench with glycine (yes - no)
  4. shearing time (4 x 20% x 30 secs - 1 x 10% x 30 secs)
  5. preclear with beads (yes - no)
  6. antibody concentration per 25 ug of starting material (2 mg - 10 mg )
  7. incubation time with beads prior to washing (10 min - 2 hr)
  8. bead type (agarose - dynal)
Along with this, I'm also going to test using Charge Switch or Qiagen PCR cleanup rather than phenol/chloroform for the initial DNA quantification step. If they are comparable, the charge switch and Qiagen columns are certainly much faster (and safer). I'm not sure how good they are with small quantities, because perhaps it could be used in the final elution too? Maybe in round 2.

2.2.1  Phenol -vs- ChargeSwitch -vs- Qiagen PCR purification for cleanup of sheared, decrosslinked DNA

While I'm waiting for reagents to come in for the factorial experiment, I'm going to test to see if it matters which cleanup kit I use in the first step where I quantify my amount of starting DNA before adding the antibody.
Wed Apr 25, 2007
I spread some lrpB-TOPO-antiExpress (lrpB) onto an amp/agar plate.
Sun Apr 29, 2007
I picked two colonies (lrpB:J1 and lrpB:J2) to grow overnight in 4 ml of LB+amp.
Mon Apr 30, 2007
I grew a 1:100 dilution of lrpB:J1 and lrpB:J2 from the overnight cultures in 25 ml of LB + amp. After 1 hr, I added 1:1000 of 1M IPTG (1 mM ).
As the cells reached an OD600 of *, I then followed the ChIP protocol for crosslinking, lysing, and shearing the chromatin. Minor modifications were: the initial cell pellet centrifigation I ran for 15 min rather than 10 min (cells then precipitated completely), after the addition of 500 ml of 2x Pallson IP buffer I incubated 5 min and only spun at 100 rpm (lysis still went to completion with slower rpm and shorter time).
For the shearing, I sheared J1 3 x 20% x 30 seconds, and I sheared J2 3 x 20% x 30 seconds.
To reverse the crosslinks, I placed 10 ml , 50 ml , and 100 ml of sheared DNA into a total volume of 125 ml (topped up with 115 ml , 75 ml , and 25 ml of H2O respectively) with 0.5 ml , 2.5 ml , and 5 ml of Proteinase K [Ambion] respectively. I did this for both J1 and J2. I made three replicates of each of these, so I'd have one replicate for each of: ChargeSwitch, Phenol, Qiagen PCR cleanup column. Total number of tubes was 2 (J1, J2) x 3 (10 ml , 50 ml , 100 ml ) x 3 (ChargeSwitch, Phenol, Qiagen) = 18 tubes. I did the dilutions to try and check the linearity of each method as the concentration of DNA changed (I'm actually concerned more with precision than accuracy, so I'd rather the slope be accurate than the intercept).
Tue May 1 19:55:55 EDT 2007
I ran 6 cleanups for all of the cleanup kits according to the manufacturer's instructions, except that for the ChargeSwitch, they recommend a 25-50 ml PCR reaction and my starting volume (besides not being from a PCR) was 125 ml . I adjusted the concentration of the initial binding purification reagent accordingly (they wanted a 1:1 ratio). I eluted each reaction into 30 ml . The yields were:
Sample DNA (ng/ul) 260/280 260/230
J1 : 10 p 21.2
J1 : 10 c 29.3
J1 : 10 Q 28.3
J2 : 10 p 28.9
J2 : 10 c 32.6
J2 : 10 Q 35.9
J1 : 50 p 123.4
J1 : 50 c 64.7
J1 : 50 Q 121.2
J2 : 50 p 79.0
J2 : 50 c 75.2
J2 : 50 Q 149.9
J1 : 100 p 663.9
J1 : 100 c 39.8
J1 : 100 Q 275.6
J2 : 100 p 1112.2
J2 : 100 c 54.8
J2 : 100 Q 299.6
(J1 = lrpB:J1 = sheared 3x; J2 = lrpB:J2 = sheared 1x; 10, 50, 100 = starting dilution of sheared DNA; (p, c, Q) = (phenol:chloroform, ChargeSwitch, Qiagen PCR cleanup column).
It's a lot easier to see what's going on with those numbers when we plot them in Figure .
Please see the pdf version for figures
Figure 2.1: The qiagen PCR cleanup kit is the only DNA cleanup method that was linear across the dilution range
Please see the pdf version for figures
Figure 2.2: The phenol:chloroform extraction lanes retained the tiny DNA fragments in the high volume purification, which may have led to the strange non-linear scaling of the measured concentrations.
Brief Conclusions:   The Qiagen concentration estimates are spot-on (Figure 2.1), but DNA purity from this kit is the worst (see table above). Not only is the Qiagen yield linear with the dilution amount, the slope is also about correct (i.e. when dilution is 1/2 the conc is 1/2). Looks like big phenol numbers are from the little pieces that the other kits remove (Figure 2.2). Also notice that although chargeswitch seems to blow, it does do a really nice job of cutting off the size of the DNA at a larger size than the other kits. I should keep this in the back of my mind, because this could be a really useful way to avoid gel purification to remove adaptors and stuff like that. They claim in the manual that the size cutoff is adjustable, so it I can push the size high enough, I might not even need to gel select the cDNA step. Last, I ordered a Qubit from Invitrogen. This is a little machine that makes using their DNA/RNA quantification kits easy (e.g. picogreen). I want to try this dilution test again, but down to much lower levels to see about the possibility of using Qiagen rather than EtOH for the final DNA cleanup after the immunoprecipitation.
In summary, it looks like the Qiagen kit is the way to go for the initial quantification of sheared DNA yield. Based on some results from Henry Lee in the Collins lab, in the next round where I try diluting further, I might quantify all of the samples with picogreen to prevent the lessen influence of mRNA in the spec readings.

2.2.2  first round ChIP optimizations: detailing the plan

Mon May 7, 2007
The factors and reasoning behind the factors was detailed above at the start of this section (i.e. section ChIP improvement round 1). Below is the table that I'm actually using to experimentally pursue this goal. The rows were randomized with the rand_perm() function in matlab.
As far as nomenclature goes, I'll be referring to the samples by their randomized order. For each sample N there is NA, NB, and NC where A = positive control sheared DNA, B = no antibody negative control, and C = antibody enriched (hopefully) sample.
Randomized 8-factor ChIP optimization 16-sample fractional factorial design
randomized order IPTG form quench shear pre anti incubate bead
1 0.01 mM 0.10% yes 4x20%x30 yes 10 mg 10min agarose
2 0.01 mM 1% yes 4x20%x30 no 2 mg 10min dynal
3 0.01 mM 0.10% yes 1x10%x30 yes 2 mg 2hr dynal
4 0.01 mM 1% no 4x20%x30 yes 2 mg 2hr agarose
5 1 mM 0.10% no 4x20%x30 yes 2 mg 10min dynal
6 1 mM 0.10% no 1x10%x30 yes 10 mg 2hr agarose
7 0.01 mM 0.10% no 1x10%x30 no 2 mg 10min agarose
8 0.01 mM 1% yes 1x10%x30 no 10 mg 2hr agarose
9 1 mM 1% yes 4x20%x30 yes 10 mg 2hr dynal
10 1 mM 0.10% yes 1x10%x30 no 10 mg 10min dynal
11 0.01 mM 0.10% no 4x20%x30 no 10 mg 2hr dynal
12 0.01 mM 1% no 1x10%x30 yes 10 mg 10min dynal
13 1 mM 0.10% yes 4x20%x30 no 2 mg 2hr agarose
14 1 mM 1% no 1x10%x30 no 2 mg 2hr dynal
15 1 mM 1% yes 1x10%x30 yes 2 mg 10min agarose
16 1 mM 1% no 4x20%x30 no 10 mg 10min agarose

2.2.3  first round ChIP optimizations: growing, shearing, lysing, sonicating

Mon May 7, 2007
I performed this step similarly to the one I did for J1 and J2 in section 2.2.1. This was the first time I had used the glycine to quench the crosslinking. I grew the cells for 3 hrs and 20 minutes before taking the 15 ml samples for crosslinking (background subtracted OD600 is in the table below). I used 750 ml of 2.5 M glycine (1/20) in the 15 ml reaction. Based off the information from that section, I cleaned up the reactions with Qiagen, and I obtained the following yields:
Sheared DNA yields from ChIP factorial optimization round 1
Sample ID ng/uL 260/280 260/230 num ml for 25 mg ml buffer for 1:10 dilution total volume OD600
ChIP 1 242.58 1.77 1.37 103 928 1031 0.406
ChIP 2 300.41 1.74 1.28 83 749 832 0.438
ChIP 3 289.23 1.97 1.63 86 778 864 0.401
ChIP 4 276.1 1.74 1.45 91 815 905 0.417
ChIP 5 300.52 1.71 1.09 83 749 832 0.368
ChIP 6 275.34 1.94 1.63 91 817 908 0.365
ChIP 7 283.41 1.99 1.55 88 794 882 0.408
ChIP 8 268.74 1.8 1.46 93 837 930 0.398
ChIP 9 268.68 1.8 1.48 93 837 930 0.367
ChIP 10 275.22 1.93 1.64 91 818 908 0.366
ChIP 11 279.2 1.82 1.62 90 806 895 0.402
ChIP 12 219.63 1.77 1.01 114 1024 1138 0.397
ChIP 13 248.9 1.8 1.51 100 904 1004 0.389
ChIP 14 257.39 1.89 1.59 97 874 971 0.377
ChIP 15 219.73 1.96 1.47 114 1024 1138 0.370
ChIP 16 305.7 1.77 1.33 82 736 818 0.344
To Do!!!  I still need to run these on a gel to check the shearing range. I also should check my previous experiments to determine how much to run on the gel
May 15, 2007
I ran the 16 factorial samples on two 1.5% agarose gels (Figure ).
Please see the pdf version for figures
Figure 2.3: 1.5% gel; 16 sheared samples; the lightly sheared samples were 1x10%x30secs. the others were 4x20%x30secs.
Brief Conclusions:   Well it's too late now (I've already run the qPCR samples), but if I had this shearing to do this factorial again, I'd make the shearing for the low-shear a little longer or stronger. The low-shear lanes have the DNA average at around 1kb or more. The high-shear lanes have the DNA average length at around 350bp.

2.2.4  first round ChIP optimizations: antibody, beads, washing, crosslink reversal

Tues May 8, 2007
Notes1: sample 3 took forever for beads to finish binding to the magnet; sample 6, 7, 10 were very viscous (I didn't notice that well with sample 7 and for that one a lot of the beads were sucked out of the tube)
Notes2: in samples 7B and 7C the beads are almost gone; 10C is a complete impenetrable ball of magnetic beads. The washes and 5 minute rotations have little effect on breaking up this ball
Notes3: I prepared the dynal beads using a similar strategy to the Young Lab protocol in Nature Protocols. 1) add N ml of beads; 2) collect with magnet; 3) add 15 x N ml of block solution (0.5% BSA in PBS); 4) repeat steps 2 and 3; 5) resuspend beads in N ml of block solution
I used the same volume of beads for dynal and agarose: 40 ml preclear and 60 ml immunoprecipitation. The washings were done in the cold room. They were done according to my original chip protocol except some of the samples were incubated with antibody and beads 2 hrs each (4 hours total) and other samples were incubated 10 min each (20 minutes total).
For the elution, I used the Young protocol (which used something similar to TE + 1% SDS) for the dynal elution. And I used the original protocol for the agarose elution. Crosslinks were reversed overnight in a water bath at 65C.

2.2.5  first round ChIP optimizations: purification of DNA products

Tues May 9, 2007
I added 1 ml of proteinase K to all 32 samples. To the dynal samples I added H2O to make them be at the same volume as the agarose samples (450 ml total). I added Tris and EDTA to the agarose samples, as per my original ChIP protocol.
I did a phenol extraction of all of the samples using Gel Phase lock (light) tubes. I did not do a second chloroform extraction. I added 1 ml of glycoblue to the EtOH precipitations. I did the EtOH precipitations in three rounds hoping that would help prevent having the DNA pellets come unstuck from the tubes. I don't think it mattered. Next time I'd just do two rounds of 16.

2.2.6  first round ChIP optimizations: preparing and testing the ChIP primers

I'm using the same 12 random primers from last time, even though I think know at least one of those primer doesn't work. 11 genes is still a fairly large negative sample for comparison with my enriched genes. The row order was randomized, as was the gene order within each row. The twelve genes chosen for lrp include:
Lrp factorial primer plate
- 1 2 3 4 5 6 7 8 9 10 11 12
A gcl mog pinO idnD yhaF nhaA aimA goaG kdtB yagG citC fruK
B yhjE gtlB stpA serC leuL serA aroP pntA metA serA5KB serA1KB livK
I made one 300 ml plate and one 100 ml plate. Both plates had primers at 2 mM concentration.
To make sure the primers still work (they're about 1yr old) and that I set up the primer plate correctly, I did a 20 ml PCR with the cheap NEB Taq and genomic DNA. I ran them out on a 2% agarose gel (Figure ).
Please see the pdf version for figures
Figure 2.4: 2% gel of the random primers and lrp primers to be used in the factorial ChIP optimization experiment. The identity of the lanes is the same as in the table Lrp factorial primer plate table above.
Brief Conclusions:   The random primer gel is a little wacko, but everything looks pretty good. mog (random lane2) and yagG (random lane 10) were wacky in the previous ChIP study too, so I'm not worried about those guys.

testing the primers with qPCR master mixes

Sat May 12, 2007
The DynamoHS SYBR green master mix from finnizymes is about 1/3 cheaper than the ABI one, so I'm going to test that one and the ABI to compare the results.
Ct values from ABI and Dynamo HS
- Ct ABI Ct Dynamo Difference
gcl 14.995 15.037 0.042
yhjE 14.542 14.902 0.360
mog 20.229 15.671 -4.558
gtlB 15.562 15.320 -0.242
pinO 19.173 16.111 -3.062
stpA 14.838 15.108 0.271
idnD 15.320 17.030 1.709
serC 14.234 14.401 0.167
yhaF 18.799 18.801 0.002
leuL 15.099 15.330 0.231
nhaA 14.921 15.169 0.248
serA 14.430 14.430 0.001
aimA 15.426 16.114 0.688
aroP 14.129 14.819 0.690
goaG 16.053 16.179 0.126
pntA 14.987 14.810 -0.177
kdtB 16.370 15.821 -0.549
metA 13.999 14.327 0.327
yagG Undetermined 36.849 Undefined
serA5KB 14.098 14.511 0.413
citC 17.118 17.105 -0.013
serA1KB 14.607 14.612 0.004
fruK 17.447 17.744 0.297
livK 14.654 14.721 0.067
Raw data in excel format.
Brief Conclusions:   The results were pretty similar besides a few outliers. One thing that you cannot see (but that I saw in the ABI software) is that the primer dimer formation as indicated by the final DNA melting curve was less with Dynamo. The dynamo also produced a little less signal over all, and it had a mildly worrying feature that after the signal saturation point was reached, the signal actually began decaying a little (the ABI master mix rises slightly after PCR saturation). The Ct values, which are in the area of the curve that really matters, were fine though.

2.2.7  first round ChIP optimizations: dynamoHS qPCR

Sun May 13, 2007
I ran samples 1-8 B and C (16 total) with all 24 genes (384 samples total) using 10 ml dynamo HS master mix, 5.5 ml H2O , 0.4 ml ROX (1x final concentration), 1.5 ml primer (2 mM stock), and 3 ml template (as in the previous experiments for the PLoS paper, I diluted the 100 TE+template with 100 ml of H2O for a final volume of 200 ml ). I made enough master mix for 10 extra rxns and ran short 19 rxns (that brings back bad memories!). I ordered a matrix multichannel electronic pipettor that should help alleviate this problem in the future. The 19 rnxs I ran short of master mix on, I made a second batch of master mix for and filled them by hand. The plate barcode was A302JWNB. As far as the 384-well plate organization goes. A1 (and all odd columns) of the qPCR plate contains A1... A12 of the lrp plate (the random row), A2 (and all even columns) of the qPCR plate contains B1... B12 of the lrp plate. For the template samples 1B... 8B was put in rows P,N,L... B; samples 1C... 8C was put in rows O,M,K... A.
Tues May 15, 2007 I ran samples 9-16 B and C (16 total). The reaction concentrations were the same as above. I made enough master mix for 30 extra rxns and did NOT run short this time. The plate barcode was A302JWNC.
Also, I did a quick skim of the qPCR data for the first eight experiments. I identified by eye the three samples that I felt were noticeably better than all of the other experiments: all had 1% formaldehyde in common (rather than the 0.1% formaldehyde).
Here is the raw data for the hsDynamo qPCR rxns factorial rnd1 samples 1-8 and factorial rnd 1 samples 9-16.

2.2.8  first round ChIP optimizations: ABI master mix qPCR

I'm trying the exact same qPCR reactions as above but with the ABI master mix that we pay about a third more for and that I used in the PLoS netinfer paper ChIP experiments. Since ROX is included in the master mix I added water in place of the 1:50 dilution of ROX in the above dynamoHS experiments.
Wed May 16, 2007
I ran the first qPCR plate with the ABI mix.
Tues May 22, 2007
I ran the second qPCR plate with the ABI mix.
Here is the raw data for the ABI qPCR rxns factorial rnd1 ABI samples 1-8 and factorial rnd 1 ABI samples 9-16.

2.2.9  Summary of first round results

Summary results from lrp factorial first round
hsDynamo qPCRABI qPCR
id factor effect pval (Lenth) effect pval (Lenth) conclusion
1 IPTG -0.1787 0.031 -0.3665 0.012 high IPTG may be slightly worse
2 formaldehyde 1.0555 0.0002 1.1803 0.0005 form. concentration is very important (1% >> 0.1%)
3 quench 0.6774 0.0007 0.5972 0.0021 quenching definitely helps
4 shear 0.0271 0.6357 -0.1646 0.13 shearing alone is not significant
5 -0.0362 0.7402 0.1825 0.1633
6 -0.2359 0.0119 0.1192 0.2048
7 -0.0492 0.9181 0.0393 0.6178
8 form/quench? -0.7273 0.0005 -0.8768 0.0011 if you have low formaldehyde - don't quench
9 form/shear? 1.0807 0.0002 1.1853 0.0005 if you have high formaldehyde - shear more
10 0.4694 0.0016 0.5356 0.0045
11 preclear -0.053 0.9784 -0.0151 0.4762
12 antibody conc -0.0532 0.9819 0.032 0.5709
13 incubation time 0.3245 0.0053 0.4424 0.0098 2hr incubation is a little better than 10 minutes
14 bead type -0.65 0.0007 -0.9791 0.0008 dynal beads aren't as good as agarose
15 - 1 - 1
Brief Conclusions:   Mon Jun 18 14:25:42 EDT 2007
I'm just going to summarize my gut feelings based on the above table.
Main effects (linear).   First, lower protein concentration might be important. Perhaps, having lrp on a high-copy plasmid with high concentration of IPTG for induction just overwhelms the genome and the lrp binds everywhere. Second, formaldehyde concentration is the single most important factor screened in this round. The 1% was much more enriching than the 0.1%. Presumably, the 0.1% is just not enough to bind everything together in those 10 minutes. When I picked the best enrichments by eye before performing the computational analyses of this data the three best I choose all had 1% formaldehyde. Third, quenching with glycine helps. Halting the crosslinking should improve the consistency and doesn't add much extra work. Fourth, shearing did not significantly effect the results. Fifth, preclearing didn't make a difference. Sixth, adding a lot of extra antibody didn't help. Seventh, 2x2hr incubations was a fair amount better than 2x10 minute incubations. Eigth, dynal was not quite as good as the agarose beads.
Combinatorial effects (nonlinear).   Unfortunately the combinatorial effects are all confounded with each other. I'm going to make my best guess at the 2-factor interactions that could explain the combinatorial interactions we see in the table above. My main guess is that since formaldehyde was the most important main effect, the combinatorial effects were probably in some way related to formaldehyde. So I'm guessing the non-linear effect number 8 is an interaction between formaldehyde/quenching. If that's the case, the effect would be interpreted as lack of quenching is beneficial when you are using 0.1% formaldehyde. This hypothesis seems pretty reasonable, because 0.1% didn't work as well as 1%, so the longer you crosslink with the low concentration, the closer you are to the shorter incubation at the high formaldehyde concentration. The second non-linear interaction I'm guessing is an interaction between shearing and formaldehyde. If so, it seems to be beneficial to shear more for higher concentrations of formaldehyde, presumably because the higher formaldehyde concentrations bind everything up together.
What I'm a gonna do.  
  1. use low IPTG (I'm also going to clone the tagged lrp into a low copy plasmid to bring the expression down even further)
  2. quench - this I'll do from here on. it takes a trivial amount of additional work. and seems to help things out
  3. keep formaldehyde concentration high. I want to explore this further in the next round with higher formaldehyde concentrations
  4. shear fairly well. seems to help with higher formaldehyde. if not, it doesn't hurt (though it does take extra time). I also want to explore the interaction with formaldehyde later in a response surface experiment
  5. don't preclear - needless waste of time
  6. antibody concentration - higher didn't matter; I'm going to try lower in the next round
  7. use faster incubation time - I know, 2 hr was significantly better, but it wasn't drastic. Maybe I should explore this in more detail to see what really matters.
  8. use dynal beads - again, I know this is in contradiction to the results, but it is much faster to use the dynal beads, and it's easier to be consistent, so I feel I'll have less noise if I go with the dynal
  9. use hsDynamo master mix from Finnizymes (distributed by NEB); about half the cost and the results are very similar

2.3  scratchNotes

Ideas for next optimization round: 1) mix beads and antibody together during the DNA clean up/quanitification; 2) use Qiagen cleanup at the end (rather than EtOH); clone into ilaria's vector with the very low copy number
Based on factorial, go with sample 15 (randomized sample 2) as the default: 0.01uM IPTG, 1% formaldehye, shear 4x20%x30, no preclear, 2 mg antibody, incubate 10 minutes, dynal beads
In first round, formaldehyde was the biggest factor. Quenching was helpful. And there was a nonlinear effect that you needed to shear more with increased formaldehyde concentration.
spec with qubit?
Fo the second round,

2.4  Second round ChIP optimizations

summarize first round here. justify second round here.

2.4.1  second round ChIP optimizations: detailing the plan

2nd round the factors are (low - high):
  1. starting DNA conc. (20 mg - 80 mg of DNA); need to grow in 50 ml flask
  2. formaldehyde (0.5% - 4%)
  3. lyse (normal - lyse in dilution buffer via sonication)
  4. shearing time (6 x 20% x 30 secs - 2 x 20% x 30 secs)
  5. antibody (0.25 ml - 2 ml )
  6. bead concentration (10 ml dynabeads - 100 ml dynabeads)
  7. wash (normal 3 salt wash + 2xTE - 2x TE only)
  8. final DNA cleanup (phenol - qiagen PCR cleanup)
For the high DNA concentration, I plan to grow 60 ml of cells and lyse 50 ml (rather than my typical 25 ml grow 15 ml lyse).
For the lyse in dilution buffer, I plan to lyse the cells in dilution buffer during sonication (i.e. let the sonication lyse the cells rather than using lysozyme and high conc. sucrose) rather than the normal chemical lysis.

2.4.2  second round ChIP optimizations: checking the sonication

Mon May 21, 2007
I was a little surprised to see how not-sheared my DNA was for the low shear factor in round 1 of the optimization (Figure 2.3). This round I'm going to check the shearing range a little in a pretest just to make sure the shearing is a little more reasonable (even though the low shear seemed to work).
Specifically, I wanted to see the effect of using no lysis and doing the sonication in IP buffer. I made two sample culture volumes. Consistant with the volumes necessary or the factorial factor of 20 ml vs 80 mg of starting DNA, I grew 15 ml and 60 ml respectively. I also tested two formaldhyde concentrations (0.5% vs 4%) and two shearing lengths (2x20%x30 secs and 6x20%x30 secs), as it was found that there was a nonlinear interaction between these in the first round, and because I'm not sure how these DNA concentrations and formaldehyde concentrations will influence the shearing size.
The six 50 ml samples were taken from individual flasks with 70 ml starting volume of culture. The six 15 ml samples were taken from one 70 ml sample and the leftover culture from the 50 ml experiments (20 ml was left in the 50 ml cultures). The cells were grown for 3hr 30 min to an OD600 of around 0.45.
I ran the following 12 samples:
sample sample vol. formaldehyde conc.chemical lysis sonication strategy
1 15 ml 0.5% no 2x20%x30sec
2 15 ml 4% no 2x20%x30sec
3 15 ml 0.5% no 6x20%x30sec
4 15 ml 4% no 6x20%x30sec
5 50 ml 0.5% no 2x20%x30sec
6 50 ml 4% no 2x20%x30sec
7 15 ml 0.5% yes 2x20%x30sec
8 15 ml 4% yes 2x20%x30sec
9 50 ml 0.5% yes 2x20%x30sec
10 50 ml 4% yes 2x20%x30sec
11 50 ml 0.5% yes 6x20%x30sec
12 50 ml 4% yes 6x20%x30sec
Tues May 22, 2007
The 12 samples were cleaned up with a Qiagen PCR cleanup kit and eluted into 30 ml of EB buffer. The yields from the nanodrop were:
Sheared DNA yields from shearing test of factorial round 2
Sample ID ng/ul 260/280 260/230
1 228.11 1.8 1.58
2 171.32 1.77 1.52
3 234.09 1.71 1.11
4 183.86 1.81 1.57
5 474.14 1.81 1.8
6 487.99 1.78 1.64
7 206.36 1.82 1.66
8 165.32 1.74 1.16
9 420.99 1.82 1.87
10 351.9 1.81 1.75
11 429.26 1.81 1.89
12 458.31 1.8 1.86
3 ml of sheared DNA was run on a 1.5% agarose gel (Figure ).
Wed May 23, 2007
I also spun down all 12 samples at max speed for 3 minutes to see if there was any more cellular junk for the sonicated only -vs- the chemical lysis + sonicated samples. None of the 12 samples had a noticable precipitate.
Please see the pdf version for figures
Figure 2.5: shearing range tests with different concentrations of DNA and formaldehyde, different shearing amounts, and different buffers. gel is a 1.5% agarose gel.
Brief Conclusions:   I like these results better than what I saw above for the round1 shearing checks (see Figure 2.5 vs the round one shearing in Figure 2.3). There isn't a huge difference between these samples, but the 6x sheared are noticably shorter. Note that the quantities of DNA in each lane aren't the same for the 50 ml samples and the 15 ml because the yields were different and I just ran 3 ml for all samples. By eye, I don't see any noticable difference between the samples that were chemically lysed and those that were lysed as part of the sonication process.

2.4.3  second round ChIP optimizations: factors

I'm pretty much using the protocol I wrote up from the last round (see section on page ). I'm just adding in the factors above.
randomized setup for factorial round2
randomized order DNAconc form lyse shear anti beadConc wash DNAclean factorialOrder
1 20ug 4% no 2x20%x30 2ug 100ul normal qiagen 3
2 80ug 0.50% yes 6x20%x30 0.25ug 10ul 2xTE only phenol 14
3 20ug 0.50% no 6x20%x30 0.25ug 100ul 2xTE only qiagen 9
4 80ug 4% yes 6x20%x30 2ug 100ul 2xTE only qiagen 16
5 80ug 4% yes 2x20%x30 2ug 10ul normal phenol 8
6 80ug 4% no 6x20%x30 0.25ug 100ul normal phenol 12
7 80ug 4% no 2x20%x30 0.25ug 10ul 2xTE only qiagen 4
8 20ug 4% yes 2x20%x30 0.25ug 100ul 2xTE only phenol 7
9 80ug 0.50% no 2x20%x30 2ug 100ul 2xTE only phenol 2
10 80ug 0.50% no 6x20%x30 2ug 10ul normal qiagen 10
11 20ug 4% yes 6x20%x30 0.25ug 10ul normal qiagen 15
12 80ug 0.50% yes 2x20%x30 0.25ug 100ul normal qiagen 6
13 20ug 0.50% yes 2x20%x30 2ug 10ul 2xTE only qiagen 5
14 20ug 4% no 6x20%x30 2ug 10ul 2xTE only phenol 11
15 20ug 0.50% no 2x20%x30 0.25ug 10ul normal phenol 1
16 20ug 0.50% yes 6x20%x30 2ug 100ul normal phenol 13
The randomized matrix of factors is available here: in excel format.

2.4.4  second round ChIP optimizations: growing, shearing, lysing, sonicating

Sun May 27, 2007
I grew the cells up as before (adding IPTG 1hr after the 1:100 incubation).
OD600 (background subtracted) and sheared DNA yields were:
OD and sheared DNA yields for lrp ChIP factorial round2
Sample ID OD600 ng/ul 260/280 260/230 10ug or 25 ug
1 0.474 152.41 1.89 1.74 131
2 0.467 348.89 1.83 1.67 143
3 0.462 172.9 1.92 1.92 116
4 0.468 373.37 1.84 1.66 134
5 0.465 383.56 1.81 1.59 130
6 0.46 378.22 1.83 1.81 132
7 0.456 334.07 1.94 1.8 150
8 0.477 183.45 1.89 1.8 109
9 0.494 372.1 1.85 1.66 134
10 0.482 480.82 1.43 0.67 104
11 0.507 253.12 1.7 0.9 79
12 0.501 335.63 1.84 1.68 149
13 0.511 313.93 1.7 0.89 64
14 0.491 181.65 1.89 1.91 110
15 0.493 184.1 1.91 1.93 109
16 0.486 175.03 1.88 1.78 114
Notice in the table above I include the amount needed to get 10 mg and 25 mg . In the original plan, I was going for 20 mg and 80 mg , but I realized that I forgot to consider that I was eluting into 1/2 of the volume after I cleaned up the sheared DNA, so in the past I've been using 10 mg NOT 20 mg of DNA (opps). It was impossible to get up to 80 mg because that wouldn't fit into my tube, so I went with the biggest number that would fit, which was 25 mg .
Note this time instead of using slightly different amounts of sheared DNA for immunoprecipitation, I just used the median volume for that concentration of DNA (i.e. 10 mg or 25 mg ). The median volume was 134 ml sheared-crosslinked DNA and 1207 ml Dilution Buffer for the 25 mg samples and 110 ml sheared-crosslinked DNA with 986 ml Dilution Buffer for the 10 mg samples.
Please see the pdf version for figures
Figure 2.6: 1.5% gel run 45 min at 120V; 16 sheared randomized samples from factorial round 2

2.4.5  second round ChIP optimizations: antibody, beads, washing, crosslink reversal

Mon May 28, 2007
everything went fine here just a simple note: 6C and 6B were slower to bind the magnet (presumably they were more viscious) than 1C or 1B.

2.4.6  second round ChIP optimizations: purification of DNA products

Tue May 29, 2007
In a change for this round, I used both phenol chloroform and the Qiagen clean. For the phenol chloroform, I added 200 ml TE and 4 ml protease K as in Lee et.al. Nature Protocols. For the Qiagen kit, I just added 4 ml protease K. I incubated for 1 hr at 45 C.

2.4.7  second round ChIP optimizations: dynamoHS qPCR

Wed May 30, 2007
I switched to 10 ml rxns to save money and because I didn't have enough master mix to use 20 ml . I hope to go back and try the 20 ml rxns to see if it makes a difference. Since the total volume of master mix was much less, I put the master mix in PCR strips rather than a reservoir.
Plate 1: A302JWNE Plate 2: A302JWNF
Both of the plates had B in the bottom-left corner. HOWEVER, in the previous experiment I used B as the no-antibody sample, whereas in this experiment it was the antibody sample.
Here is the raw data for the hsDynamo factorial rnd2 samples 1-8 and factorial rnd 2 samples 9-16.

2.4.8  Summary of second round results

Click for little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see factorialBatchRound2.m for the script).
Second round ChIP factorial results
hsDynamo qPCR
id factor effect pvals
1 amt starting DNA -0.045 0.1259
2 formaldehyde 0.1516 0.0049
3 lyse -0.2299 0.0013
4 shear 0.1599 0.0036
5 0.0173 0.7927
6 -0.0533 0.1818
7 0.0426 0.1094
8 -0.3404 0.0004
9 -0.0261 0.3011
10 0.0817 0.0382
11 antibody conc 0.3085 0.0006
12 bead conc 0.3593 0.0004
13 wash -0.1931 0.0013
14 final DNA cleanup -0.022 0.9478
15 -0.2114 0.0019
Brief Conclusions:   Mon Jun 18 18:46:11 EDT 2007 I'm just going to summarize my gut feelings based on the above table.
Main effects (linear).   If anything, lysing is not helpful (removing this will save at least an hour). Again formaldehyde is important. Here, it is not as strong an effect as last time. Perhaps 4% is a big overshoot. I definitely need to fine tune this concentration. More shearing was helpful alone this time, but did not seem to have an effect in combination with formaldehyde this time. I still think it would be useful to optimize these two together. More antibody was helpful. More beads was helpful. The normal washing was better than washing with TE alone.
What I'm a gonna do after round2.  
  1. use the original amount of starting DNA - increasing the amount change help anything, and slows things down because you have to grow more and use bigger tubes
  2. back to 1% formaldehyde - I need to do a more in depth study of the formaldehyde concentration combined with changing the shearing amounts
  3. don't lyse chemically - let the sonicator do it for you; saves a lot of time and some pipetting
  4. shear fairly well - again needs to tried on its own with different formaldehyde concentrations; I want the minimal shearing time that will give the best results (since shearing is very time-consuming and labor-intensive)
  5. antibody concentration middle - last time higher didn't matter. this time, lower did matter; I'm going to explore this further with a response surface
  6. bead concentration - higher is better; will optimize with antibody conc in a response surface
  7. don't wash too much - washing is slow, labor-intensive, and uses tons of tips; For now, I'm only going to do the two TE washes. I'm going to explore this latter in more detail. I realize this was a significant effect, but the effect size was not so large large that removal of the washes will cause the proceedure to fail
  8. use Qiagen PCR purification for final DNA cleanup - had no effect; plus the results with the Qubit and the Qiagen purifications showed they were incredibly consistent down to very small amounts of DNA (see Figure ).

2.5  Response surface 1: beads vs antibody

Given the results from factorial round 2, I decided to optimize the values of the antibody and bead using the method of steepest decent.

2.5.1  preparing the sheared chromatin

Wed Jun 13, 2007
I'm just going to prepare a batch of sheared DNA/chromatin to optimize the bead and antibody concentrations. I'm not going to prepare a separate sample for each concentration beads and antibody to try. Rather, one tube of sheared DNA is enough for 4 samples, so I'm just going to make 3 tubes of sheared DNA, which is enough to try 12 different samples.
The protocol is basically the ChIP protocol with a couple modifications using what I learned in round 2. The modifications are: stay with 1% formaldehyde, don't lyse with chemicals (put cells in Dilution buffer and let the sonication pop them open). I'm also going to do the washes with TE two times rather than using all of the salt buffers for reasons I mentioned above (see section 2.4.8 on page pageref).
The sheared chromatin (4x30secx20%) for all three samples was run out on an agarose gel (Figure ).
Please see the pdf version for figures
Figure 2.7: Sheared chromatin on agarose gel. This sheared DNA will be used for the response surface of beads/antibody.
Brief Conclusions:   So far so, good. As it should be - I've sheared cells for ChIP about a million times now....

2.5.2  testing ability of Qiagen PCR prep to work with minute quantities of DNA

Thu Jun 14 19:42:30 EDT 2007
In factorial round 2, I found no significant difference between phenol:chloroform + EtOH vs Qiagen PCR purification. I wanted to dig into this further with the current batch of sheared chromatin, because the Qubit just arrived in the mail. The Qubit is a fluorometer from Invitrogen built just for quantifying DNA, RNA, and protein using Invitrogen's dyes. The dyes have a few important characteristics. First, the DNA dye is very specific for DNA not RNA, so you measure DNA concentrations almost independently of the amount of RNA in the tube (same holds in the other direction for the RNA dye). This specificity is a nice feature, because I can see if I still have RNA in my sample despite having put RNAse cocktail into my sample prior to sonication (e.g. maybe the sonicator deactivates the RNAse). The second key feature of their dyes is that they allow you to measure very small quantities of DNA (down to 10 pg/ml ). I still don't know how much DNA I pull down with the ChIP proceedure, but I do know it is smaller than I can measure with the nanodrop (I've tried with the nanodrop and only get rubbish). The Qubit might allow me to quantify this DNA.
The goal for this second is to take the three sheared samples I described in the previous section and to dilute them with TE to several different concentrations. Then I'll clean them up and quantify with the nanodrop and the Qubit.
Qubit and Nanodrop readings from Qiagen purified sheared DNA
Sample ml sheared DNA Qubit (ng/ml ) Nanodrop (ng/ml ) Nanodrop (260/280) Nanodrop (260/130)
1 100 377 454.45 1.81 1.87
1a 50 158 168.28 1.82 1.81
1b 10 26.2 41.84 1.73 1.33
1c 5 14.6 29.52 1.67 1.05
1d 1 3.06 42.98 1.56 0.63
1e 0.5 1.174 12.35 1.42 0.71
1f 0.1 0.228 37.13 1.52 0.55
1g 0.01 0.0339
2 100 297 356.3 1.8 1.78
2a 50 140 167.78 1.78 1.61
2b 10 23.4 44.55 1.72 1.19
2c 5 11.68 25.99 1.57 1.08
2d 1 3.02 15.58 1.4 0.71
2e 0.5 1.454 16.44 1.28 0.96
2f 0.1 0.316 26.69 1.48 0.63
2g 0.01 0.0347 13.09 1.37 0.72
3 100 303 333.13 1.81 1.82
3a 50 127 150.47 1.79 1.78
3b 10 21.6 44.56 1.73 1.06
3c 5 12.1 27.11 1.7 1.03
3d 1 1.974 12.44 1.65 0.8
3e 0.5 1.216 13.31 1.51 0.8
3f 0.1 0.252 12.72 1.47 0.66
3g 0.01 0.0312 12.78 1.41 0.69
The above table is easier to see as a graph (Figure ).
Please see the pdf version for figures
Figure 2.8: fill in
Brief Conclusions:   I certainly wasn't expecting my graph to look this good (Figure 2.8). Maybe if I'd just diluted a known concentration of DNA, I'd expect a graph this nice. But this is a Qiagen PCR Purification of different quantities of sonicated ChIP starting material. Not only is the Qubit linear, the Qiagen cleanup must be extremely consistent across different concentrations of DNA as well. Notice that the nanodrop begins to overestimate the quantity of DNA at around 20-40 ng/ml . According to this figure, the nanodrop doesn't really work at all below 20 ng/ml . In general the nanodrop is about 10-20% higher then the Qubit readings. Perhaps this reflects the RNA remaining in the sample? Overall, this experiment went really well, and it increases my confidence to totally switch to Qiagen PCR purification kits for cleaning up the final ChIP DNA rather than slow, laborious, and hazardous phenol:chloroform extraction.

2.5.3  optimization of bead and antibody concentrations

Now that we've learned a little about what matters in the ChIP protocol, we can start to push forwards and optimize what matters. Our previous factorial experiments pointed out the factors and allowed us to speed up the protocol by removing things that didn't matter from the protocol. What we want to know now is can we get more enrichment. The first step is the optimization of bead and antibody concentrations, which were found to be important in factorial round 2.
The stuff I describe now comes from the books Statistics for experimenters, 2nd Edition pages 489-537 and from BOX on QUALITY and DISCOVERY pages 146-169.
Using least squares, I determined the parameters for altering bead and antibody concentrations using steepest ascent (antibody parameter = 0.1542; bead parameter = 0.1797). I rounded and scaled these parameters to 15 and 18 respectively and calculated four pairs of values for antibody and beads.
BEGIN NOTE
Ilaria figured out how to move right in the table in Figure 12.4 in Statistics for Experimenters. For the example in the book the second row is:
x = [(-13)/28] ·0.75 ·[0.875/0.875]
the third row is:
x = [(-8)/28] ·0.75 ·[0.375/0.875]
The first row is determined by the unit size chosen by the experimenter.
END NOTE
The values for antibody and beads determined by this method were:
antibody concentration (mg ) 0.25 1.25 2.25 3.25
bead concentration (ml ) 10 70 130 190
For each reaction, I used 100 ml of sample and 900 ml of dilution buffer. I ran two sets of these four concentrations. The main set was with dynal beads (the parameters actually came from the factorial analysis that used dynal beads only). The second set used agarose beads. I did this to give the agarose beads a second chance, as they performed a little better than the dynal in the first round. If they prove much better in this analysis, perhaps they're worth the extra hassle.

PCR rxns agarose vs dynal

I ran 10 ml PCR rxns for the dynal and the agarose beads. The average enrichment and standard deviation for each bead/antibody concentration set is shown in Figure . Bundling all of the genes into one mean and standard deviation creates a lot of noise (hence the big error bars), as some genes enrich much more than other genes.
Here is the raw qPCR data for the agarose and dynal rxns
Please see the pdf version for figures
Figure 2.9: ChIP enrichment of positive control genes relative to random genes for bead/antibody concentrations determined via steepest assent.
Brief Conclusions:   The bead/antibody concentrations were determined using the data from factorial round 2 where dynal beads were used. Therefore, it's a little unfair to compare the agarose and dynal beads at these concentrations, since the agarose might have had completely different concentrations. Nevertheless, dynal does seem to have less noise and is more consistent than the agarose beads.
The response curve for the dynal bead is quite nice, looks just like it does in the textbook. We can see how the enrichment increases as we pile on more beads and antibody. One thing we don't know from this plot (since I didn't carry it out far enough) is whether or not it saturates and if it saturates does the enrichment start to decline when more bead is added.

PCR rxns old primers vs new primers

I ordered a new set of random primers and a new set of lrp primers for a couple reasons: 1) the previous primers I used were over a year old (though they have been in the freezer the whole time and only freeze-thawed 3x); 2) I want to be absolutely sure that what I'm seeing is not primer specific. In the new set of lrp primers, I picked a few of the primers from before and a few new ones. For the ones that were included from before, I redesigned a new primer for them (i.e. the primer will bind a slightly different location). The primer plate with the new random and new lrp primers was 2 mM as before and was laid out as follows:
Here is the raw qPCR data for the new lrp primers
Lrp factorial primer plate2
- 1 2 3 4 5 6 7 8 9 10 11 12
A ynhG apaG ypdB sanA ybaO arpB infC hybA tdcA mviM ygfZ ycfX
B leuA ilvH lrp dadA oppA aroA livK serA ilvA lysU fimA stpA
I tested the primers on sheared DNA, and ran them out on a gel (Figure ). I think these gels are really only useful for making sure you didn't do something completely idiotic like mix the primers incorrectly. This check keeps you from wasting an entire qPCR plate, because none of the primers worked. However, it is important to do a second check once you have the qPCR results. On qPCR results for ChIP, genes that consistently have a Ct > = 30 when amplifying the immunoprecipitated DNA should probably be eliminated from further analysis. They just don't behave properly. Good Ct values will be from 20-28 or so.
Please see the pdf version for figures
Figure 2.10: new lrp, random, cysB, and lexA primers amplified from sheared DNA
The same enriched DNA was used for the new primers as had been used with the old lrp primers in the section above. The results comparing the new primers with the old primers are shown in Figure .
Please see the pdf version for figures
Figure 2.11: comparison of a newly synthesized set of lrp target primers relative to the performance with the old set I've been using up until now
Brief Conclusions:   The results are pretty dang similar between the two primer sets. Very nice... The only annoying thing was that very few of the primers I designed for lrp worked properly (i.e. the Ct values were too high, indicating that they weren't binding the DNA efficiently enough to be reliable). All 12 of the random genes worked beautifully though. Might be a good idea to make a plate3 that is a composite of 1 and 2. Would contain all N primers that I know work well: livK, serA, serC, pntA, dadA, yhjE, gtlB(?).

2.6  third round ChIP optimizations

Given the data in the antibody:bead response surface, I want to do one final factorial optimization of the second half of the ChIP protocol. The second half entails everything that occurs after the shearing.
For this final screening, I'm going to go with a full factorial design to allow easy interpretation of the results, particularly regarding any potential interactions we might find.
The factors are:
  1. antibody (1.25 - 2.25 mg )
  2. beads (70 - 130 ml )
  3. silicon tubes (no - yes)
  4. LiCl wash (no - yes)
The antibody and bead concentrations came from the previous response surface results above. Lot's of folks claim siliconized tubes work better to reduce noise, so I gave that a shot. And I also brought back the LiCl wash, which was significant in a previous round. I wanted to check it in this more rigorous full factorial setup to hopefully get a better gauge on the importance of the salt washes.
In order to look potentially fit a nonlinear surface to this data, I also included 3 replicates of the middle points, 2 samples below the factorial points, and 3 samples above the factorial points (see the table below to figure out what I'm talking about).
ChIP factorial 3rd round setup
ID silicon bead anti liclWash randomized order
1 no 70 1.25 no 8
2 yes 70 1.25 no 2
3 no 130 1.25 no 3
4 yes 130 1.25 no 23
5 no 70 2.25 no 22
6 yes 70 2.25 no 13
7 no 130 2.25 no 19
8 yes 130 2.25 no 1
9 no 70 1.25 yes 6
10 yes 70 1.25 yes 10
11 no 130 1.25 yes 9
12 yes 130 1.25 yes 15
13 no 70 2.25 yes 14
14 yes 70 2.25 yes 24
15 no 130 2.25 yes 16
16 yes 130 2.25 yes 21
17 yes 100 1.75 yes 20
18 yes 100 1.75 yes 5
19 yes 100 1.75 yes 17
20 yes 10 0.25 yes 18
21 yes 40 0.75 yes 11
22 yes 150 2.75 yes 12
23 yes 190 3.25 yes 4
24 yes 250 4.25 yes 7

2.6.1  third round ChIP optimizations: checking the sonication

Wed June 20, 2007
I grew up 6 samples in 30 ml of LB. Sonicated 4x30secx20%. No chemical lysis. I could spec the cultures, because the plate reader wasn't working.
Thur June 21, 2007
I cleaned and spec'd the six samples (volume was the typical 50 ml ):
Sample DNA (ng/ul) 260/280 260/230
1 428.7
2 454.9
3 457.2
4 431.4 1.79 1.73
5 396.3 1.80 1.73
6 441.5 1.80 1.83
Yields were slightly higher than my normal readings (normally I get 200-300 ng/ml ). I ran the sheared samples on a gel (much later Figure ).
Please see the pdf version for figures
Figure 2.12: Six sheared samples on 1% agarose gel.
Fri June 22, 2007
I ran the binding and wash steps for all 24 samples detailed in the table above. The samples were run in the randomized order, pulling from each of the 6 sheared samples until their was no sample left (e.g. randomized samples 1-4 used sheared sample 1, 5-8 used sample 2, etc...). The samples were placed at 65C overnite to remove the crosslinks.
Sat June 23, 2007
I cleaned up all 48 samples (24 samples without antibody, 24 samples with antibody) using Qiagen PCR cleanup columns. I eluted into 100 ml EB.
I ran 2 qPCR plates from on the purified DNA (this was the first 16 samples).
Mon June 25, 2007
I ran the final plate of the experiment.
Here is qPCR plate 1 of factorial 3, qPCR plate 2 of factorial 3, and qPCR plate 3 of factorial 3

Results from the 3rd round factorial

Figure .
Click for little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see factorialBatchRound3.m the script).
Please see the pdf version for figures
Figure 2.13: Enrichment (the z-axis displayed via color) certainly is dependent on antibody and bead concentrations. There appears to be some saturation (too much DNA?).

2.7  Preparations for future factors

I want to make a low-copy version of the AntiXpress-tagged lrp protein to see if bring the expression down even lower has an effect.

2.7.1  Cloning lrpB into a low-copy plasmid

To get the expression of LrpB down further, I want to change the copy number (which is currently very high) to low copy number. 
Forward primer
----------------------------------------
     XhoI   promoter -35
ATAT CTCGAG TGTTGACAATTAATCATCCGGCTCGTAT

Reverse primer
----------------------------------------
     AvrII
TTAA CCTAGG ATTTGTCCTACTCAGGAGAGCGTTC
Tue Jun 19, 2007
I'm cloning the gene into the plasmid Ilaria uses (derived from the one U. Alon uses a lot). It has a very low-copy number (6 per cell?).
I amplified the lrp from the lrpB-TOPO plasmid that I've been using for ChIP. I added XhoI and AvrII using the primers above. I used 1 ml plasmid and 2 ml primer (from 10 mM stock) in the 30 ml reaction. I used the Finnizymes Phusion Taq and 30 cycles. Yields were:
Sample DNA (ng/ul) 260/280 260/230
lrp-lowcopy 34.6
I ran 5 ml of the PCR rxn on a gel (Figure ) to verify that it was the correct size (I was a little worried it would difficult to amplify over the 2' structure of the transcription terminator).
Please see the pdf version for figures
Figure 2.14: gel of lrp PCR prior to cloning into low-copy plasmid
I digested the DNA and the vector (obtained from Ilaria) using 2 ml NEBuffer2, 2 ml BSA, 15 ml template, 0.5 ml AvrII and 0.5 ml XhoI. I ran the digestions for 20 min followed by 10 min heat deactivation at 65C. I gel purified the digested vector using a Qiagen Gel Purification column, eluting into 30 ml . The gel was stained with Sybr Safe and I didn't take a picture of it, so that transformation efficiencies would be high (don't want the UV light messing up my DNA).
Unfortunately, I made 2 big mistakes with this and it didn't work the first time. First, I loaded my digested PCR into a gel, rather than cleaning it up with a PCR purification. I caught my mistake before I started the gel, so I sucked up the cut PCR product from the gel well with a pipettor. I cleaned up the DNA I sucked out of the gel well using a Qiagen PCR purification kit. I don't think this is an efficient way to clean up your DNA though :). Then in a bigger mistake, I assumed the plasmid was amp resistant and plated the transformation on an amp plate. The plasmid was actually kanomycin resistant. The negative control had zero colonies, but the positive control did have a few colonies (strange, but I'm not the first person to find this http://www.bio.net/bionet/mm/methods/2005-March/099322.html). I picked one of the colonies just for the hell-of-it, and it didn't grow in Kan.
Wed Jun 20, 2007
Since I only cut 15 ml of the PCR rxn yesterday, I had 15 ml left that I could cut today. I ran the same ligation as yesterday (but not for the vector, because I still had some gel-purified left from yesterday). I cleaned up the digestion using a Qiagen PCR purification kit, eluting into 30 ml . Then I ran the following ligation: 5 ml PCR digestion, 8 ml cut vector, 2 ml T4 buffer, 1 ml T4 ligase, 4 ml H2O . I incubated 10 min at RT; I did not heat deactivate. Rather, I directly transformed 2 ml into DH5a competent cells.

2.8  forth round ChIP optimizations: chromatin concentration

The previous two rounds clearly show that bead and antibody concentrations are important. With more beads and more antibody and you get more enrichment. It's not clear what the surface of displaying the interaction between these two variables will look like. However, in the previous round I appears that I need a heck of a lot of beads and antibody to get the best enrichment. I want to fit this surface describing the interaction between beads and antibody, but it would be expensive to do given the current amount of beads that my results suggest I should use.
The question then becomes are we saturating the beads and antibody with DNA? If so, we should be able to drop the DNA concentration and move the quantities of bead and antibody down accordingly. If this is the case, then the ChIP protocol becomes much more flexible. Knowing the interaction between these three, you can always maximize enrichment by choosing the optimum concentrations of beads and antibody for each amount of DNA. By increasing the amount of DNA, you can (presumably) grab more DNA which might be better for cloning the enriched DNA. By decreasing the amount of sheared chromatin DNA, you can lower costs, particularly if you are doing sensitive qPCR reactions to verify enrichment.

2.8.1  the plan

To test this, I'm running the following full factorial with three factors.
ChIP factorial 4th round setup
ID bead (ml ) anti (mg ) sheared DNA (ml ) randomized order
6 130 0.75 100 1
7 40 2.25 100 2
2 130 0.75 25 3
8 130 2.25 100 4
5 40 0.75 100 5
3 40 2.25 25 6
4 130 2.25 25 7
1 130 0.75 25 8

2.8.2  grow, lyse, shear

Mon Jul 9, 2007
I grew 6 samples as in the previous factorial experiment. I grew them to 0.5 OD600. I lysed them via sonication.
Tues Jul 10, 2007
I cleaned and spec'd the six samples (volume was the typical 50 ml ):
Sample DNA (ng/ul) 260/280 260/230
1 207.8
2 226.7
3 212.3
4 235.4
5 198.6
6 192.5

2.8.3  immunoprecipitation

Tues Jul 10, 2007
I proceeded according to the factorial setup above using the randomized order. I used 900 ml of dilution buffer for all rxns whether I used 25 ml of sheared DNA or 100 ml of sheared DNA. Since there were only 8 conditions, I only used samples 1 and 2 of sheared DNA (immunoprecipitations with randomIds 1-4 used sheared sample 1 and 5-8 used sheared sample 2).

2.8.4  elute DNA and qPCR

Wed Jul 11, 2007
I cleaned up the samples with a Qiagen PCR purification kit. Unfortunately, I dropped a bunch of the tubes onto my bench as I was closing the tubes for the final time. The only samples that noticibly lost a little volume were 2c and 7c. However, since I'm looking at relative ratios, it shouldn't make a big deal.
I ran 10 ml qPCR using hsDynamo and the lrp plate 2 with random primer set 2.
Here is the raw data for the qPCR rxns factorial rnd4 all 8 samples.

2.8.5  Summary of fourth round results

Click for a little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see factorialBatchRound4.m).
Please see the pdf version for figures
Figure 2.15: figure summarizing results of the 4 factorial experiment. Note that I've shown the forth dimension (DNA concentration) by putting the points right next to each other. The low DNA concentration is on the left and the high DNA concentration is on the right. This makes it appear that the low DNA sample has slightly less beads than the high DNA sample, this isn't the case. I just change the amounts slightly on purpose so the datapoints wouldn't overlap. For an alternative view see Figure
Please see the pdf version for figures
Figure 2.16: figure summarizing results of the 4 factorial experiment. this figure is plotted in 3d for an alternative view to Figure 2.15 for livK.
Brief Conclusions:   Very nice when your hypothesis turns out to be right. In Figure 2.15 notice that all of the points on the left (low DNA) are more enriched than those on the right, and in Figure 2.16 the top points (top as in the points on the top if you image this thing is a cube; that is DNA conc = 100 ml ) are pretty much always less enriched than the corresponding point right below them (i.e. DNA = 25 ml ).
So it does look like we'll be able to scale the amount of antibody and bead down but using much smaller amounts of sheared chromatin starting out. I don't know yet how this will be affected by going to low copy number plasmid (i.e. will I need more sheared chromatin?). I also still need to see the affect of bead and antibody saturation (i.e. do things start to go bad when the sample is overwhelmed with beads and antibody?
Since I have 4 sheared DNA sample left, I think I want to drop the concentration even lower. Where I stand now, I should use 25 ml of sheared chromatin (appx 2.5 mg of DNA) with 100 ml of beads and 2 ml of antibody. I'd like to get the optimum bead amount down to 50 ml and then fill out the entire matrix of antibody to bead.

2.9  fifth round ChIP optimizations: finishing up the bead antibody surface

Thu Sep 20 13:21:05 EDT 2007
After a long break (for postdoc interviews and to publish a couple computational papers), it's time to finish this thing up. Given the recent confirmation that our chromatin was saturating the beads and the antibody, we can tone things down and figure out the surface in more detail without using 10 gallons of expensive dynal beads.

2.9.1  fifth round (bead/antibody surface) experiments

Wed Sep 19, 2007
I grew 50 ml lrpB in one 250 ml flask (originally I wanted to keep with my standard 25 ml in 125 ml flask, but all of the flasks were dirty). I grew the cells from a 1:100 dilution for 3 hr 30 minutes to an OD600 of 0.456; I added 1% formaldehyde, incubated 10 min, and quenched with glycine. Sonication was 4x30secx20%. 100 ml of the sheared chromatin was left overnite at 65C to reverse crosslinks.
Thur Sep 20, 2007
I cleaned up the crosslinked-reversed DNA with a Qiagen PCR purification kit. I eluted into 50 ml of EB, yields were:
Sample DNA (ng/ul) 260/280 260/230
Sample 1 169.9
Sample 2 188.1
Sample 3 231.4
I ran 2.5 ml of all three samples on a 1% agarose gel (Figure ).
Please see the pdf version for figures
Figure 2.17: Sheared chromatin for ChIP. Average length is around 400 bp.
Fri Sep 21, 2007
Based on the previous rounds and a little intuition, I decided generate the antibody/bead surface with 16 samples in a 4x4 matrix of antibody (0.75, 1.5, 2.25, 4.5 ml ) and beads (40, 70, 100, 200 ml ). As always, I randomized the order of the samples excel file of table:
sampleID randID antibody (ml ) bead (ml )
1 16 0.75 40
2 6 1.5 40
3 14 2.25 40
4 12 4.5 40
5 1 0.75 70
6 13 1.5 70
7 5 2.25 70
8 9 4.5 70
9 15 0.75 100
10 11 1.5 100
11 10 2.25 100
12 3 4.5 100
13 7 0.75 200
14 2 1.5 200
15 8 2.25 200
16 4 4.5 200
I used 17.5 ml of chromatin sample 3 (231.4 ng/ml = 2.025 mg /sample) for each precipitation. The preciptations were run with a 30 min antibody incubation and a 30 min bead incubation. The beads were washed 2x and eluted into 210 ml Elution buffer (from the Lee et al. Nature Protocols paper).
I also calculated the amount of DNA from the previous rounds of ChIP where I varied the antibody. When I got sloppy earlier, because I didn't think chromatin concentration mattered too much, I began just always using 100 ml of chromatin. Now that I know that chromatin concentration affects the saturation point of the beads and the antibody, I can actually use bead/antibody data at those different chromatin concentrations to include chromatin concentration as a third variable in my model (nice when sloppiness is actually useful). I wrote a matlab script to get the enrichment values along with their corrresponding chromatin, antibody, and bead concentrations (see the function combineDNA_Anti_Bead for the script). Note including the previous run which is at around 2 mg per sample, I've used 2.65, 10.6, 16.3, and 21.2 mg of chromatin in different runs). For most of these samples, I actually pulled from one of several sheared chromatin samples for each precipitation reaction. I just took the average chromatin concentration from all of the sheared chromatin to calculate the amount for the experiment, since I didn't actually specify in the experiment which of the chromatin samples I used for each reaction.
Mon Oct 1, 2007
There's been a bit of a delay in running the qPCR on those bead/antibody samples, because our 12-channel small volume pipettor was broken. Turns out it was irrepairably broken, so I ordered two new ones from biohit a mechanical (m10) and an electronic (proline). The proline arrived today, so I'm going to run the first plate and give the new pipettor a try. I to make an lrp-superplate containing all of the positive control genes that I'd used that I knew worked (i.e. the primer pair got a decent Ct value in the mid-to-low twenties). I used the random_primer_plate2 primers for the random genes (just as I've been doing for most of these experiments). I randomly ordered the 12 positive primers, except that I left the positives from the previous bead/antibody runs in the same location on the plate (just in case there was something magical about the locations they were in). I also included the serA1kb primer which is 1kb away from the known serA site. And I included 3 replicates of the livK from the second plate, so that for one gene I'd have a triplicate qPCR technical replicate for all of the ChIP samples. The primers with a "2" after the gene name are from the second primer plate I ordered from IDT.
Lrp factorial superplate (2 mM )
- 1 2 3 4 5 6 7 8 9 10 11 12
A ynhG apaG ypdB sanA ybaO arpB infC hybA tdcA mviM ygfZ ycfX
B livK serC serA1kb dadA2 livK2b pntA livK2 serA2 yhjE gtlB serA livK2c
I ran samples with random ids from 1-8 on the first plate. Unfortunately, either the new proline pipettor doesn't seal well on the tips or I didn't give the pipettor enough time to charge before I started, because by the end the pipettor was aspirating very unevenly if at all. Because of that, there were a few wells on my 384-well plate that certainly didn't have the correct volume. So I think I'm going to repeat these qPCR experiments when the mechanical m10 pipettor arrives and hope for the best.
Here is the raw data for the qPCR run bead antibody samples 1-8 proline pipettor and the matlab script I used to partially analyze them is here: factorialBeadAntiSurface1, though I didn't do too much analysis because I wasn't happy with the quality of the data. From my quick scan, the general trend of the data looked about right.
The data for livK were by far the best because, as you can see in the primer table above, I placed three technical replicates for the livK samples (livK2, livK2b, livK2c).
Wed Oct 3, 2007
I received a new low volume pipettor an mLine (m10) mechanical pipettor. I decided to use the to test samples 9-16 of this antibody-bead surface. I'm skipping trying to clean up samples 1-8 for the moment, because I only have enough of 1-8 to give it one more shot. So I'd prefer to try the new pipettor on samples 9-16 where I have to chances left. Learning from the success of including technical replicates on my plate. I abandoned the lrp-super primer plate from the last experiment and created yet another primer plate. This primer plate contained only dadA, serA, livK, and pntA each in triplicate. dadA, serA, and livK were all from the second lrp plate I had synthesized by IDT. The column ordering was randomized with the matlab randperm() function:
Lrp factorial technical replicate plate (2 mM )
- 1 2 3 4 5 6 7 8 9 10 11 12
A ynhG apaG ypdB sanA ybaO arpB infC hybA tdcA mviM ygfZ ycfX
B dadA livK serA serA dadA pntA dadA pntA serA livK pntA livK
The new pipettor proved to be a godsend, and thankfully, I don't think I'll have to worry too much about poor tip seal on my low volume multichannel. The pipetting was quite accurate. I liked the pipettor enough to write a review of the m10 on IzziD. I'll delay discussion of the qPCR results until after I've run samples 1-8 using the new mLine multichannel.
Wed Oct 4, 2007
I ran samples 1-8 using the m10 pipettor. No problems.

2.9.2  Summary of fifth round (bead/antibody surface) results

Fri Oct 5, 2007
Click for a little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see factorialBeadAntiSurface2.m).
Please see the pdf version for figures
Figure 2.18: Contour plots from round 5
Brief Conclusions:   Figure 2.18 shows contour plots fit using the median qPCR technical replicate enrichment value for dadA, livK, and serA. Note that in creating this figure, I cheated and removed one point (bead = 70 ml , antibody = 4.5 ml ) that was vastly more enriched than other samples around it. I'm guessing this was some kinda outlier. I'd need a few more replicates at that point before I'd believe that 70 ml of bead with 4.5 ml of antibody mysteriously does better than slight deviations from those values. You can see from the contour plots that there seems to be a narrow window for livK and serA where you get optimal enrichment. I think this is due more to my sampling. I was expecting (hoping?) to see that at the extreme upper-right corning the values would taper off a little like they do (for example from some sorta bead or antibody saturation). However, I only have antibody samples at 2.25 and 4.5 and I think the small size of the bright red region is due to the cubic fit not having any samples from 2.25 to 4.5. To clean up this surface, I'd need to fill in a few of the holes, and hopefully the bright red section would increase. I just realized I forgot to add a colorbar to these charts. But really I just want to show the trend, rather than the absolute enrichment anyways.
So to conclude, I think that the surface - particularly for livK and serA - looks pretty promising and as I expected it to look. The only negative thing, I dropped the surface to 2 mg of DNA based on the experiment above (see section 2.4.8 page pageref) where I showed I could get more enrichment with less bead and antibody by using less chromatin DNA (presumably the chromatin was saturating my bead and antibody). However, the maximal regions in Figure 2.18 don't require that much less bead or antibody than the maximal values in Figure 2.13 where I used much more DNA.
I was hoping to use the combined bead/antibody/DNA values to fit a nice model showing the relation ship between these three variables using a 2nd order taylor series model (combineDNA_Anti_Bead.m is the matlab script to at least grab all of the values into one set of variables). So far the relationship isn't really making sense. It works at one level of DNA concentration, but really looks weird as I make 3d surfaces at different DNA concentrations. Ilaria set up all of the modeling stuff. I think the problem is overparameterization given the sparsity of antibody bead concentrations tested at all DNA concentrations besided the 2 mg used in this round 5.

2.10  Sanity check: again, does lowing DNA concentration really allow us to lower the amount of bead and antibody

Fri Oct 5, 2007
After section 2.4.8, it seemed clear that I'd be able to drop down the DNA concentration lower and lower until M ml bead and N ml antibody would allow maximal enrichment (that is I could lower the DNA and likewise lower the amount of antibody and bead need to get maximal enrichment of that amount of DNA. all I needed to do was to figure out the mathematical relationship between the three variables). However after finishing section 2.9.2, I'm having my doubts (see the brief conclusions in that section for my reasoning).
So I'm going to look at DNA concentration for a second time. In this experiment, I'm going to try a wide-range of DNA concentrations to see if my hypothesis from section 2.4.8 strengthens or weakens.
I'm going to use four chromatin DNA concentrations (10 mg , 2 mg , 0.4 mg , and 0.08 mg ) across two combinations of bead/antibody (50 ml /1 ml and 100 ml /2 ml ). I used sample 2 (188.1 ng/ml ) from section 2.9.1 as the chromatin sample. The randomized experimental design was (excel file):
sampleId randID bead antibody DNA (mg )
6 1 2 100 0.4
7 2 1 50 0.08
2 3 2 100 10
8 4 2 100 0.08
5 5 1 50 0.4
3 6 1 50 2
4 7 2 100 2
1 8 1 50 10
Sat Oct 6, 2007
I cleaned up the reversed-crosslinked enriched DNA using Qiagen PCR purifications. I used the lrp-technical-plate from round 5. However, in that plate one of the dadA replicates didn't work (presumably I forgot to put the primers in that particular colume). I added primer from my stock to that broken well so that dadA would have all three of its samples for calculating it's median qPCR-based enrichment value.
Click for a little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see DNA_conc_data.m).
Please see the pdf version for figures
Figure 2.19: Sheared chromatin for four DNA concentrations using two combinations of bead and antibody
Brief Conclusions:   The highest two concentrations in Figure 2.19 are the same DNA concentrations I used for the data in Figures 2.15 and 2.16. The two experiments are in agreement. I get better enrichment with 2 mg of DNA than I do with 10 mg . However, from Figures 2.15 and 2.16, I assumed the hypothesis that the beads and antibody were being saturated by too much DNA had become more believable. Using that hypothesis, I extrapololated that I should be able to keep dropping the amount of DNA until I could get a maximal enrichment with a smaller amount of bead and antibody (to drastically cut the costs of ChIP). The new results in Figure 2.19 don't support this extrapolation at all. After I get below 2 mg , the results become very inconsistent across my genes. With my extrapolation, you'd also think that the red and blue points (high bead/antibody -vs- low) would start to converge and eventually intersect; the blue points would be on top at low DNA and the red points would be on top at high DNA. Instead 100 ml beat / 2 ml antibody is pretty much always better than the lower concentrations.
Either there is some minimal threshold amount of bead/antibody or else this idea of DNA saturation is not completely true (or maybe not true at all?). If the saturation idea is not true at all, I don't have any way to explain why the 2 mg has now repeatedly out performed 10 mg of DNA.

2.11  Does the volume of the precipitation matter?

Mon Oct 8, 2007
Given the previous results suggesting that lowering DNA chromatin concentration wouldn't allow me to likewise use less bead and antibody and still obtain maximal enrichment (section 2.10), I decided to try one final thing to see if I can reach my goal of being able to control bead and antibody concentrations. Perhaps the reason more bead works better is because the beads then represent a larger precentage of the volume in my tube. I usually use 1 ml in each precipitation reaction. With 200 ml of beads, 20% of my reaction is beads, but with 50 ml of beads, only 5% of my reaction is beads. The manual that Invitrogen sends with their beads also suggests minimizing your volume. However, all of the papers I've read doing ChIP with dynal beads use at least the volume I use - often more.
Similar to the previous DNA concentration test (section 2.10), I'm going to try four levels of my feature of interest using both 1 ml :50 ml and 2 ml :100 ml of bead:antibody respectively. The randomized experimental design is (excel file of experimental design):
- randID sheared chromatin (2 ug) dilution buffer antibody bead
5 1 24 151 1 50
8 2 24 13.5 2 100
7 3 24 13.5 1 50
3 4 24 426 1 50
4 5 24 426 2 100
2 6 24 976 2 100
6 7 24 151 2 100
1 8 24 976 1 50
Note that when I prepared the beads, I resuspended them all (both 50 ml and 100 ml beads) into 100 ml of PBS. So the final volume in each tube is the volume of dilution buffer plus 24 ml of sheared chromatin plus 100 ml of bead/PBS. For the two TE washes for each sample, I used the volume of TE corresponding to dilution buffer plus 100 ml (e.g. for sample randID=1 I used 251 ml of TE for each wash; for randID=2 I used 113.5 ml TE for each wash).
Tues Oct 9, 2007
I qPCR'd the 8 samples.
Click for a little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see volume_bead_antibody.m).
Please see the pdf version for figures
Figure 2.20: Enrichment when using a series of different volumes for the precipitation reaction
Brief Conclusions:   Initially when I saw the resulting enrichment -vs- volume plot, I thought I just had noise and that volume didn't matter (Figure 2.20). However, if you look ignor the second point from the left (volume = 251 ml ). The low bead/antibody samples (blue) follow a pretty darn straight line that maxes out at the most concentrated volume. The high bead/antibody samples (red) aren't as clean. They initially do better as you concentrate them more, but then they perform worse as you concentrate them further. Above (section 2.10), when I was scanning the DNA concentrations, I assumed that at some point the red and blue samples should intersect with the blue becoming more enriched - perhaps that's what I'm seeing here? However, I haven't forgotten that the second point from the left might not be an outlyer at all and all I'm hoping for here could just be a bunch of rubbish. I think there is a strong enough hint of a signal in this data to inspire me to give this a second try to replicate. Unfortunately, I'm outta sample, so first I need to prepare some sheared chromatin...

2.12  Sanity check: does the volume of the precipitation matter?

Sun Oct 14, 2007
My results in section 2.11 were decent enough that I want to explore further if the reaction volume is important for ChIP. Unfortunately, I'm outta fresh sheared chromatin. So I just started an LrpB (i.e. AntiXpress-tagged Lrp) overnite culture.
Mon Oct 15, 2007
I grew up 3x50 ml culture in 250 mml baffled flasks using a 1:100 dilution of the overnite. After one-hour I added 0.01 mM of IPTG. After 3 hr and 30 minutes, the ODs for the three cultures were in the correct range: 0.53, 0.518, and 0.485; I took two 15 ml samples from each flask and added the standard 1% formaldehyde for 10 min followed by quenching with glycine and two washes in ice-cold PBS.
Mon Oct 16, 2007
The six samples were sonicated 4x20%x30sec. 100 ml sample was crosslink-reversed and purified with a Qiagen PCR purification column. 5 ml of each sample was run on an agarose to verify the correct shearing range (Figure ). The yields (elution into 50 ml ) were:
Sample DNA (ng/ul) 260/280 260/230
1 127.5
2 139.9
3 126.4
4 125.4
5 129.5
6 121.2
Please see the pdf version for figures
Figure 2.21: Six samples where sheared, crosslink reversed, and cleaned up. 5 ml of each (appx 600 ng) were run in each lane. The average shearing length is around the expected 300-400 bp range.
Brief Conclusions:   As with my formaldehyde/shearing samples (section page ), the yields were about half of my usual values. The only thing I can think of is that I believe I recieved a new tube of RNAse Cocktail [Ambion] prior to the formaldhyde/shearing samples. Perhaps, the old cocktail was getting old and didn't remove all of the RNA?
Tues Oct 16, 2007
I ran a randomized enrichment volume scan using the exact values as the previous attempt (section 2.11), but the order was rerandomized. The design was (excel file of design):
- randID sheared chromatin (2 ug) dilution buffer antibody bead
6 1 31 151 2 100
7 2 31 13.5 1 50
2 3 31 976 2 100
8 4 31 13.5 2 100
5 5 31 151 1 50
1 6 31 976 1 50
3 7 31 426 1 50
4 8 31 426 2 100
For each sample I used 31 ml of sheared chromatin sample 1 (127.5 ng/ml ) from the table above.
Wed Oct 17, 2007
I cleaned up the crosslink-reversed samples using Qiagen PCR purification columns. I ran a qPCR on all the samples.
Click for a little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see volume_bead_antibody.m). This data was combined with the previous replicate (section 2.11).
Please see the pdf version for figures
Figure 2.22: Combined data from the two volume scan replicates. All points are the average of two replicates except the two points at 251 ml . These each (i.e. 1:50 and 2:100 bead:antibody) had a screwy outlier ( > 2x log units away from all other points) that I removed.
Brief Conclusions:   The faint signal hinting at a linear increase in enrichment with a decrease in the rxn volume is much cleared when the data from this new replicate is averaged with the previous replicate (compare Figure 2.20 and with the averaged data in Figure 2.22). For dadA and serA there is a clear linear increase in the enrichment. The livK behaviour is a little different; it appears to increase in enrichment followed by a small decay as the volume reaches the smallest levels (the pink points are noisy, but I think the cyan points are a good representitive of what I'm refering to - it's almost like a candy cane shape). The other gene I'm following (pntA) that isn't in this plot, also displays a similar shape to livK.
Unfortunately, the points at 251 ml only have one replicate, because the other sample for each of the two volumes was a pretty extreme outlier (it was > 2x the log distance away). I threw out these outliers, but the 251 ml point remains quite noisy and I'd really like to clean it up.
Thur Oct 18, 2007
In an effort to try and clean up the 251 ml points, I ran three replicates at that point for both the 1:50 and the 2:100 antibody:bead combinations. The design was (excel file of design):
id randID antibody bead volume
3 1 1 50 251
2 2 2 100 251
1 3 1 50 251
6 4 2 100 251
4 5 2 100 251
5 6 1 50 251
I cleaned up the enriched DNA with a Qiagen PCR purification column.
Sat Oct 20, 2007
I ran the qPCR plate, but unfortunately, the data was just a mess. Lots of reactions failed and those that worked had much higher Ct values than I'm used to. Later, I discovered that the primer plate was almost empty, so perhaps I wasn't pipetting my primer concentrations accurately. Either way, this data is not terribly useful and so for the time being the previous plot (Figure 2.22) will have to do.
The raw data for these failed and poor PCR rxns is here: 10_20_07_4volumeChecksReplicate251x3.

2.13  Making the bead/antibody surface using a small reaction volume

Ok, now back to where we were before I developed an obsession for trying to get enrichment using less beads and antibody rather than just using the huge amounts of antibody and bead that seemed optimal. I really want to be able to get optimal growth using a maximum of 50 ml of Dynal Beads. Initially, I thought lowering the amount of DNA was the key. But further tests suggested that this was only of limited help (Figure 2.19. However, it turned out that lowering the reaction volume looked like a more promising way to increase enrichment (Figure 2.22).
One really nice implication from the increase enrichment with smaller volume is that enrichment step might be really easy to adapt to a 96-well format. Towards that, I'm going to use a total volume of 200 ml (the volume of a standard PCR strip and 96-well PCR plate) for this bead/antibody surface.
Thur Oct 18, 2007
I'll be using 29 ml of sheared sample 2 from sheared chromatin samples in section 2.12. So each enrichment will contain 29 ml sheared chromatin (2 mg ), 71 ml of dilution buffer, and 100 ml of beads. Different amounts of beads will all be concentrated and resuspended in a final volume of 100 ml PBS + 0.5%BSA to maintain a consistent 100 ml of beads. The TE washes will be 200 ml .
I tested 12 total combinations. A 3x3 matrix of 1,2,5 ml antibody combined with 50, 100, and 200 ml of bead. In addition I added an extra 3 points in other spots of interest. The design was (excel file of design):
id randId antibody bead
8 1 2 200
2 2 2 50
6 3 4 100
12 4 3 150
10 5 3 150
4 6 1 100
5 7 2 100
3 8 4 50
1 9 1 50
7 10 1 200
9 11 4 200
11 12 3 150
Sun Oct 21, 2007
I ran the two qPCR plates. Shortly before running the qPCR I realized I had only tiny amounts of primer left in my primer plate (lrpTechReplicate). Thankfully, I had a second aliquot in the freezer, which I thawed and fixed the problem I'd earlier had when I forgot to add the final dadA primer to its third technical replicate well in the primer plate. The two PCR plates were run back-to-back with 12 samples in each plate. Click for little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see factorialBeadAntiSurface3.m for the script).
Please see the pdf version for figures
Figure 2.23: Bead/Antibody enrichment surface for high and low volume samples. The A) high volume samples panel is taken from Figure 2.18 but with the addition of the actual datapoints used to fit the contour.
Please see the pdf version for figures
Figure 2.24: Bead/Antibody enrichment surface for high and low volume samples. The A) high volume samples panel is taken from Figure 2.18 but with the addition of the actual datapoints used to fit the contour. This plot is exactly the same data as Figure 2.23 except that both the high volume and the low volume are plotted using the same scale for the colorbar.
Brief Conclusions:   When you compare the previous bead/antibody surface (Figure 2.24A) with the new low volume version (Figure 2.24B), it is quite clear that I achieved my goal of being able to get my original maximum enrichment using only 50 ml of antibody. However, what I wasn't expecting to find was that I could get even more enrichment by still using the high amounts of bead/antibody at the lower volume. The plots for dadA and serA get about 3x more enrichment at the low volume, and the enrichment is on a log scale so the old max enrichment for the two genes was around two-fold more than random, it is now around five-and-a-half fold more enriched than random. In these low volume plots, it appears that antibody is starting to play a more dominant role.
To keep costs down, I'm going to do the remaining optimizations using 3 ml of antibody and 100 ml of bead for each reaction. I think that will provide a nice balance between maximal enrichment and cost.

2.14  Plate -vs- tube

Thur Oct 25, 2007
Now that the volume is down below 200 ml , we can attempt to enrich our targets using a plates or pcr-strips rather than with 1.5 ml tubes. Doing so would make running hundreds of samples much more tractable.
I purchased a 96-well Dynal magnet from Invitrogen (Dynal MPC-9600). Rather than just switching straight-away, I want to try the enrichment with tubes and with PCR-strips to see if there are any differences. For the tubes, I'm running the reaction just like before. For the PCR strips, I'm not rotating the samples. Rather I'm using the magnet to mix beads, as recommended in the MPC-9600 manual.
I used 32 ml of the sheared DNA (2 mg ) from section 2.12, 100 ml of bead, and 58 ml of dilution buffer (for a total rxn volume of 190 ml ). Samples 1-4 were the old tube way. Samples 5-8 were the new PCR-strip/96-well way. I changed buffers and eluted the samples in the PCR strips using a multichannel pipettor to verify that it isn't hard to work with the beads using a multichannel (I did the initial bead allocation with a single-channel though to prevent wasting beads).
Fri Oct 26 20:10:24 EDT 2007
I cleaned up the crosslink reversed samples with a Qiagen PCR purification kit. When I did the purification, I forgot to elute the samples and left them sitting in EB buffer for quite a while (10-20 min); hopefully, that doesn't mess things up. I ran the qPCR on all 8 samples. Click for little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see plate_vs_tube.m for the script).
The mean enrichment values for dadA, livK, serA, and pntA were 0.2760, 0.7373, 0.2934, and 0.1101 respectively. Although these were all positive, they were all much lower than the mean values from the bead/antibody surface in the previous section (1.1058, 2.0953, 0.9377, 0.5430). A two-tailed t-test between the plate and the tubes was not significant for any of the four tested genes pvals=(0.1681, 0.0935, 0.1490, and 0.4282), suggesting that there was not difference between the plate and the tubes (if you just look at the mean and ignor the noise, the plate did slightly better).
Brief Conclusions:   It looks like there is not difference between the tube and the plate which is good - high-throughput here I come! However, the whole-experiment is tainted by the fact that the enrichment values for all four genes dropped by so much. I think the answer lies in the Ct values. The Ct values were much lower for this experiment than for all previous ChIP experiments I've run - ever. The mean Ct was 17 for this experiment whereas for the previous experiment is was 22. Since all four genes had positive enrichment, clearly I'm picking up signal, but it looks like the signal is masked by background noise. I'd guess I'm picking up DNA from somewhere? Next time I run a ChIP experiment, I must make fresh dilution buffer, fresh elution buffer, fresh 0.5% BSA/PBS buffer, use a new box of qPCR master mix, clean my bench and pipettes throughly with DNA-away, and use a new source of DNA-free water.
How much this background signal interfered with my t-test is not clear. However, for now I'm going to move onward and assume moving to a plate doesn't really matter. The major difference between the two was that I mixed with rotation for the tubes. But with a volume of 200 ml the rotation really doesn't mix much anyways. Sunday I hope to make the formaldhehyde/shearing surface that I thought I'd make about a month ago...

2.15  Making the formaldehyde/shearing surface

Tues Oct 2, 2007
The last surface I want to make is the optimize formaldehyde concentration and shearing amount. My previous factorials strongly showed the influence of formaldehyde and hinted that there might be some interaction with shearing. Now I'm going to test this in more detail. I have two goals:
  1. can we increase our ChIP enrichment still further by optimizing shearing and formaldehyde
  2. shearing is still the least fun part of the protocol, so I'd like to the least amount of shearing that will give me an optimal result
I initially thought I'd use a 4×4 matrix of formaldhyde × shearing. However, I'm more concerned with formaldehyde, so I decided to do a 4x3 matrix. I added an addition 4 points at the far corners to try and help determine the boundards of the formaldehyde/shearing plot. The four formaldehyde values were: 0.5, 1, 2, 4%. The three shearing values were 1, 2, and 3x30secx20% power. The boundaries four points were a two-by-two matrix of formaldehyde=(0.1% and 8%) and shearing (1x30secx20% power and 6x30secx20% power).
The final set of 16 experiments was:
id randID formaldehyde shearing formaldhyde % ml formaldehyde in 15ml total vol
5 1 0.5 3 0.1 0.041
12 2 4 4 0.5 0.2
1 3 0.5 2 1 0.405
7 4 2 3 2 0.81
3 5 2 2 4 1.62
4 6 4 2 8 3.24
15 7 0.1 6
11 8 2 4
8 9 4 3
16 10 8 6
2 11 1 2
10 12 1 4
9 13 0.5 4
13 14 0.1 1
14 15 8 1
6 16 1 3
Tues Oct 3, 2007
I cleaned (Qiagen PCR; elute 50 ml ) and spec'd the samples and ran all 16 onto a 1% agarose gel for 40 min at 110V (Figure ). The spec values are (raw spec data):
Sample ID Date Time ng/ul 260/280 260/230
1 10/3/07 3:00 PM 147.56 1.89 2.01
2 10/3/07 3:01 PM 140.23 1.86 1.93
3 10/3/07 3:01 PM 131.66 1.86 1.91
4 10/3/07 3:02 PM 120.89 1.87 1.94
5 10/3/07 3:02 PM 119.08 1.87 2.06
6 10/3/07 3:05 PM 104.88 1.83 1.95
7 10/3/07 3:09 PM 112 1.86 1.96
8 10/3/07 3:10 PM 118.57 1.82 1.93
9 10/3/07 3:11 PM 105.18 1.84 1.93
10 10/3/07 3:12 PM 91.33 1.78 1.74
11 10/3/07 3:13 PM 100.53 1.85 2.03
12 10/3/07 3:14 PM 107.44 1.82 1.88
13 10/3/07 3:14 PM 108.62 1.82 1.93
14 10/3/07 3:15 PM 90.67 1.82 1.81
15 10/3/07 3:16 PM 31.02 1.78 1.73
16 10/3/07 3:16 PM 102.05 1.82 1.96
test1 10/3/07 3:19 PM 220.22 1.82 1.64
test2 10/3/07 3:20 PM 217.05 1.81 1.53
test3 10/3/07 3:20 PM 223.82 1.78 1.37
Please see the pdf version for figures
Figure 2.25: Sheared chromatin for different formaldehyde and bead concentrations
Brief Conclusions:   The yields of about 100 ng/ml are about half of my normal yields. I'm not sure why, as the OD I grew the cells to was the same as normal. The three test spec readings above were run to make sure that the lower yields weren't resulting from errors with the plate reader (the tests seemed fine). Note random sample 15 had an extremely low yield. This was the high formaldehyde (8%) low shear sample.
Sun Oct 28, 2007
After a long break to sort out all kinds of stuff relating to the bead/antibody/DNA amount/and reaction volume, I'm finally read ready to enrich these guys. Using the small volume, 96-well approach developed above, I enriched each sample using 2.5 ml antibody and 100 ml of Dynal beads. For samples 1-14,16, I used the average of those sample DNA concentrations to estimate the amount of sheared chromatin to use for 2 mg (35 ml ), the remaining 55 ml was dilution buffer. For the weak sample 15, I just used the maximum of 90 ml chromatin. When I washed, I resuspended the 100 ml of bead into 90 ml 0.5% BSA/PBS to reduce the reaction volume a bit. The total volume (not counting the antibody) for each of the samples was then 180 ml .
Antibody and bead incubations were both 40 minutes. And elution was for 15 min at 65C in a water bath in 180 ml of Dynal elution buffer. The samples were placed at 65C overnite to remove crosslinks.
Mon Oct 29, 2007
I added 4 ml proteinase K, incubated for 1 hr at 55C, cleaned up all 16 samples with Qiagen PCR purification columns, and eluted into 100 ml EB buffer.
I ran a qPCR on samples 1-8.
Tues Oct 30, 2007
I ran a qPCR on samples 9-16.
A plot of the results is shown in Figure .
Click for a little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see form_shear_original.m for the script; the two qPCR files are 10_29_07_shear_formaldehydePlate1.txt and 10_30_07_shear_formaldehydePlate2.txt).
Please see the pdf version for figures
Figure 2.26: It seems that lower formaldehyde concentrations are enriching better.
Brief Conclusions:   These results contradict my earlier formaldehyde result from the first factorial experiment where the 1% formaldehyde produced a boost of around 1-log more enrichment than the 0.1% formaldehyde (see section 2.2.9 on page pageref). The lowest formaldehyde/shearing combination performed best. Kinda sucks because I feel stupid for checking all of those points all over the place when the bottom left point is the best! (Ideally the best point would be in the middle so I could see the performance of the parameter space around the optima).
However if this result proves true, it does have a couple nice implications: 1) I don't have to shear as much, which is good because shearing takes forever and is horribly boring; 2) I can use lower formaldehyde which opens the possibility that it'd be easier to cut the crosslinked DNA.

2.15.1  adding more low formaldehyde concentration data points

Because the results in the previous section contradict my earlier factorial result, I decide to try and fill in the form/shearing matrix with some additional low-concentration formaldehyde and low-shearing datapoints.
The randomized experimental design is (excel file of experimental design):
low-concentration formaldehyde shear surface design
sampleId randomId formaldehyde shearing
1 6 0.033 1
2 3 0.1 1
3 7 0.3 1
4 8 0.6 1
5 5 0.033 2
6 1 0.1 2
7 2 0.3 2
8 4 0.6 2
formaldehyde percent formaldehyde in 15 ml
0.6 240 ul
0.3 120 ul
0.1 40 ul
0.0333 14 ul
Wed Oct 31, 2007
started overnite culture of LrpB
Thur Nov 1, 2007
I grew four cultures in LB in 50 ml baffled flasks from a 1:100 dilution of the overnite culture for 3hr 30 min to an background subtracted OD600 of: 0.469, 0.479, 0.474, 0.485 (0.01 mM IPTG was added after 1hr of growth). Two samples were taken from each of the four cultures to make the eight total samples in the experimental design above. Formaldehyde concentrations and shear where done according to the experimental design table above. 25 ml H2O and 5 ml proteinase K was added to each 100 ml of each sample, prior to placing them at 65C overnite to reverse crosslinks. The remaining 900 ml of sheared chromatin was placed at -20C.
Fri Nov 2, 2007
I cleaned up the overnite crosslink-reversed samples using a Qiagen PCR purification kit, and ran them on a 1.5% agarose gel (Figure ). Yields were:
overnite crosslink reverse with proteinase K
Sample DNA (ng/ul) 260/280 260/230
1 125.8
1 138.0
2 142.3
3 172.4
4 203.5
5 144.1
6 149.1
7 169.9
8 149.5
Please see the pdf version for figures
Figure 2.27: Crosslink for all 8 samples were reversed at 65C with proteinase K overnite (A) or for 1 hr (B).
I used the median DNA concentration value from the table above (149.3) to calculate the amount of chromatin to use for each IP (27 ml chromatin, 53 ml dilution buffer). I used 100 ml of beads washed 1x in 0.5% BSA/PBS and resuspended in 100 ml of 0.5% BSA/PBS.
I recorded the time necessary to prepare the beads in 96-well format: 15 minutes.
Sat Nov 3, 2007
I ran the qPCR for the 8 new formaldehyde/shearing concentrations and I combined the new results with the previous results (for 24 total formaldhyde/shearing combinations). A plot of the results is shown in Figure . Click for a little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see form_shear.m for the script; the newest plate is 11_03_07_shear_formaldehyde_plate3.txt and the previous qPCR files are 10_29_07_shear_formaldehydePlate1. txt and 10_30_07_shear_formaldehydePlate2.txt).
Please see the pdf version for figures
Figure 2.28: Further confirmation that the lower formaldehyde concentrations are performing best
Brief Conclusions:   With the addition of the additional 8 samples (Figure 2.28), we have further support that the lower-formaldehyde and shearing amounts are the best performing - contradicting the earlier factorial result.

2.15.2  Do I really need to reverse the crosslinks overnite to quantify the amount of sheared chromatin

Given that the 96-well format allows my to do the enrichment stuff in less than two hours, if I didn't have to reverse the crosslinks overnite to quantify the amount of sheared DNA, I could do the entire protocol in almost a day (at the end of which I was reverse the crosslinks on the enriched samples, and clean them up + qPCR them the following morning = 1.5 days).
To test this, I took the eight samples from the low-concentration formaldhyde experiment above and I took an extra 100 ml to quantify where I incubated them at 65C with the proteinase K for only an hour before I cleaned them up with a Qiagen PCR purification kit. I also took the normal 100 ml sample and reversed the crosslinks the typically overnight way with proteinase K. The yields for both methods are below (all samples were eluted into 50 ml of EB buffer):
Thur Nov 1, 2007
1 hr crosslink reverse with proteinase K
Sample DNA (ng/ul) 260/280 260/230
1 125.8
2 113.5
3 127.8
4 181.5
5 123.5
6 119.4
7 138.7
8 144.3
Fri Nov 2, 2007
overnite crosslink reverse with proteinase K
Sample DNA (ng/ul) 260/280 260/230
1 138.0
2 142.3
3 172.4
4 203.5
5 144.1
6 149.1
7 169.9
8 149.5
This the perfect kinda data for a paired t-test (I thought I'd live my whole life doing only unpaired t-tests). Comparing the 1hr with the overnite: pval = 7.8443e-04. With unpaired t-test pval = 0.0426; So it seems like there is a definite difference between overnight and 1 hr crosslink reversal for DNA quantification. However, the means - 134 and 159 for 1 hr and overnite respectively - only differ by 15%.
Brief Conclusions:   Given the decent robustness to changes in chromatin amount (e.g. 4-fold drop does affect things a bit, but 2-fold doesn't really change enrichment much), it might be possible to skip the DNA concentration entirely after a certain type of sample at a particular OD has already been run, you could just use the previous quantification values and things should be fine.
However for new samples or just to be safe, a 1 hr crosslink removal with proteinase K only differs from an overnite removal by 15%. 15% over or under is not going to make a noticable difference in the ChIP protocol, so I'm just going to switch over to using a 1 hr crosslink reversal for the DNA quantification step. This change will allow me to run the entire protocol (besides the final qPCR) in a day. It I really wanted to be closer to my target chromatin amount, I suppose it would be better to scale the 1 hr value by 1.15, but I'm not going to bother. Finally, the shearing range looks the same for the overnight and 1 hr crosslink reversals (Figure 2.27), so for checking the shearing range 1 hr looks fine as well.

2.16  testing the optimized protocol on a transcription factor besides Lrp

It's time for the real test: how well does our optimized protocol work on interactions it wasn't optimized for.
Sat Nov 3, 2007
I started overnite cultures of LrpB, PdhR, and FecI in LB.
Sun Nov 4, 2007
I started 9 cultures (3 of each) from 1:100 dilution of the overnites into 50 ml LB in a 250 mlbaffled flask. After one hour, I added 0.01 mM IPTG. I ran the latest version of the protocol using the single-day grow, lyse, shear, quantify, IP. This single-day protocol is possible because the IP is much faster in 96-well format and because the DNA quantification with 1 hr crosslink removal is almost the same as with overnite.
I timed everything to get an estimate of how long things would take.
The (excel file of experimental design for these 3 TFs with 6 replicates).
After 3 hr 10 minutes the samples were dense enough to shearing:
randCultureID gene OD600
1 fecI 0.489
2 pdhR 0.633
3 fecI 0.467
4 pdhR 0.638
5 fecI 0.476
6 lrp 0.628
7 pdhR 0.545
8 lrp 0.549
9 lrp 0.534
I took two 15 ml samples from each culture (18 total) into 15 ml centrifuge tubes. I used 40 ml of formaldehyde (0.1%), quenched with 750 ml glycine, washed 2x in PBS and sheared each sample with the Branson Digital Sonifier 250 for 2x20%x30sec.
I then did a 1 hr proteinase K (5 ml ) crosslink reversal at 65C for one hour. I spec'd all 18 samples with the nanodrop:
Random Sample ID Date Time ng/ul 260/280 260/230
1 11/4/07 6:14 PM 217.84 1.78 1.5
2 11/4/07 6:14 PM 202.72 1.67 0.95
3 11/4/07 6:15 PM 198.96 1.77 1.35
4 11/4/07 6:15 PM 202.71 1.79 1.39
5 11/4/07 6:16 PM 116.76 1.61 0.97
6 11/4/07 6:17 PM 243.05 1.72 1.16
7 11/4/07 6:17 PM 216.92 1.78 1.51
8 11/4/07 6:17 PM 202.43 1.75 1.34
9 11/4/07 6:18 PM 180.97 1.77 1.36
10 11/4/07 6:18 PM 222.51 1.81 1.6
11 11/4/07 6:19 PM 189.39 1.76 1.34
12 11/4/07 6:19 PM 196.88 1.78 1.38
13 11/4/07 6:20 PM 192.58 1.75 1.31
14 11/4/07 6:20 PM 180.55 1.77 1.33
15 11/4/07 6:20 PM 194.76 1.78 1.36
16 11/4/07 6:21 PM 176.63 1.8 1.55
17 11/4/07 6:21 PM 198.2 1.77 1.41
18 11/4/07 6:22 PM 152.04 1.79 1.65
The median values for fecI, pdhR, and lrp respectively were: 199.8, 196.5, and 195.8 ng/ml . I used 20 ml of each for the enrichment combined with 60 ml dilution buffer.
Timings:
Step Labor time Total time Description
Growth 45 min 3 hr 10 min
Crosslinking 1 hr 45 min 1 hr 45 min
Shearing 1 hr 15 min 1 hr 15 min
Quantification crosslink reverse 15 min 1 hr 15 min
Quantification cleanup/spec 60 min 60 min
Bead prep 8 min 8 min
antibody incubation 10 min 40 min
bead incubation 5 min 40 min
2x TE wash 10 min 10 min
final DNA cleanup protK 10 min 1 hr min
final DNA cleanup Qiagen purificiation 1 hr 1 hr
qPCR 45 min 2 hr
total protocol time: 11hr 5 min on day one; 4hr day two

2.16.1  primer plates for the three transcription factors

Mon Nov 5, 2007
I need some primer plates to run the qPCR reactions. In order to compare with the original protocol and results that I generated with the PLoS paper, I'm going to use the original set of random genes not the newer set. I decided to use 14 primer pairs for each transcription factor.
PdhR primer plate
- 1 2 3 4 5 6 7
A gcl mog pinO idnD yhaF nhaA aceE
B aimA goaG kdtB yagG citC fruK empty
Lrp primer plate
- 1 2 3 4 5 6 7
A gcl mog pinO idnD yhaF nhaA serA
B aimA goaG kdtB yagG citC fruK livK
FecI primer plate
- 1 2 3 4 5 6 7
A gcl mog pinO idnD yhaF nhaA fecABCDE
B aimA goaG kdtB yagG citC fruK fecIR
I used my typical concentration of 2 mM for each plate and made 400 ml of each primer pair in TE.

2.16.2  qPCR of targets of the three transcription factors

Mon Nov 5, 2007
I ran 16 samples on two qPCR plates. Unfortunately, I screwed up the ordering a bit and put the wrong primers in for a few samples which ruined them (I could run again later with correct primers).
In total I ran 5xFecI, 3xPdhR, and 5xLrp. Upon analyzing these samples, I decided to quit and not finish the remainder, as clearly, the results for PdhR and FecI (and even Lrp for that matter) were not as good as the original results for the unoptimized protocol. The serA and fecIR primers were also screwed up (I must have added them to the plate wrong). I also never bothered to run the sheared DNA on a gel to check shearing range.
Info to obtain the raw data for these make-me-unhappy results. Click for a little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see lrp_new_way.m fecI_new_way.m and pdhR_new_way.m for the scripts; the qPCR data is in the same directory files: fecI_new_way_qPCR.txt lrp_new_way_qPCR.txt pdhR_new_way_qPCR.txt
transcription factor log(enrichment) old protocol log(enrichment) new protocol
FecI 2.67 0.35
Lrp 3.13 2.2
PdhR 1.47 1.77
Brief Conclusions:   Mon Nov 19 19:01:03 EST 2007
Clearly this was almost the worst possible result. This whole time I'd been optimizing with Lrp alone, I was always a little afraid my optimizations would be Lrp specific. But I assumed/hoped the process was pretty general and the optimizations would at least partially apply to other transcription factors (or at least not make them worse).
I went into a brief experimenter's depression after these results, which is why I'm adding these experiments to this notebook about 2 weeks after I did them. The thing about the results that most surprised me was how well I'd done with the original protocol. I do remember that I took a couple of decisions along the way that were purposely negative (e.g. use dynal beads instead of agarose; shorter incubations) in order to speed up the protocol. However, I figured that the subsequent response surface optimizations of bead/antibody formaldehye/shear would compensate for those losses and in the end I'd have a protocol that is good or better than the original but with much less labor and much less time required per sample.
I have acheived the goal of higher throughput (throughput is now 10x the original given the 1.5 day protocol and the 96-well format). But at what cost? The one positive thing I could find about this negative result was that at least I knew which factors I had purposely chosen the less optimal state for in the factorial screens. I decided to rescreen those with two transcription factors to try and see if I can make the optimizations less transcription factor specific and to try and raise the overall enrichment for known targets relative to random targets.

2.17  a factorial with two TFs: removing the lrp specific optimizations

Given the sad result in the previous section where all of my optimizations have still left me with worse performance than my original protocol, I need to determine where I went wrong over the course of all of these experiments.
First let's state the accomplishments:
  1. the protocol is much faster (as little as 1.5 days compared with the original 5-7 days)
  2. the throughput is much higher (32 samples/day compared with the original 3.6 samples/day)
Seconds let's state the bad news:
  1. somewhere along the way in my pursuit of a faster/leaner/higher throughput protocol, the performance took a hit

2.17.1  strategy to regain the lost performance

Along the way, I purposely chose a few factor states that decreased performance. I did this when the decrease in performance resulted in a major reduction in protocol time or ease. My hope was that the subsequent response surface optimizations would bring the protocol up to at least as good as the original. Alas, this doesn't seem to be the case.
When I began preparing this data for publication, I made a composite excel table of all the factors I'd tested and the effect size and p-value of each factor. Now, I'm going to select a few of the most important factors where I chose the worst factor state, and I'm going to screen them again in a factorial experiment with both Lrp AND FecI (so I'll really be running two factorial experiments). I'm purposely NOT including PdhR, so I have a final TF to use for cross-validation. The factorial summarizing table is available in excel format; I didn't include it in this text, because it was too wide to fit.
The factors I decided to test were:   formaldehyde (0.1% - 1%), incubation time (overnite - 2 hr), bead type (dynal - agarose), high salt wash (yes - no). Using the knowledge gained from the shearing/formaldehyde surfaces, I'm used 2x20%x30sec for the 0.1% formaldehyde and 4x20%x30sec for the 1% formaldehyde. I also chose not to add in all of the salt washes; since I'd already shown the LiCl salt wash to not be important, I chose to only add back in the high salt wash (500 mM NaCl, 20 mM Tris, 2 mM EDTA, 1% Triton X-100, 0.1% SDS). For the agarose work, I'll work in 1.5 ml tubes with 500 ml volume for the dilution and washes. For the dynal work, I'll stick with the 96-well PCR-strip based method using the 180 ml dilution and washes from my previous optimizations. Finally, for the agarose beads I'll use the original amount of 10 mg sheared chromatin (since I never optimized the chromatin amount with agarose beads), and for the dynal beads I'll stick with the 2 mg optimized sheared chromatin value.
I ran these 4 factorials in a 8-run fractional factorial design. The design was randomized for the two TFs. excel file of the design. Here's the design:
fractional factorial design for Lrp and FecI
randID ID gene form incubation time bead type high salt wash
1 15 fecI 0.1 overnight:2hr dynal no
2 10 fecI 1 40min:40min agarose yes
3 4 lrp 1 overnight:2hr agarose no
4 16 fecI 1 overnight:2hr dynal yes
5 1 lrp 0.1 40min:40min agarose no
6 6 lrp 1 40min:40min dynal no
7 5 lrp 0.1 40min:40min dynal yes
8 13 fecI 0.1 40min:40min dynal yes
9 14 fecI 1 40min:40min dynal no
10 2 lrp 1 40min:40min agarose yes
11 8 lrp 1 overnight:2hr dynal yes
12 7 lrp 0.1 overnight:2hr dynal no
13 9 fecI 0.1 40min:40min agarose no
14 11 fecI 0.1 overnight:2hr agarose yes
15 12 fecI 1 overnight:2hr agarose no
16 3 lrp 0.1 overnight:2hr agarose yes

2.17.2  running the factorial experimental design

Tue Nov 6, 2007
started overnites of lrp, fecI, lrp
Wed Nov 7, 2007
grew 1:100 dilution from overnite of Lrp and FecI into two 250 ml flasks of each (4 flasks total).
After 3 hr the two lrp samples were 0.694 and 0.658 (OD600), but the fecI samples were only 0.112 and 0.119; once I saw this, I remembered (and verified by digging through this lab book) that I'd had this problem when working on the PLoS paper ChIP samples (see the first chapter Chromatin Immunoprecipitation of this lab notebook to find other examples of slow fecI growth). So I processed the lrp samples first. After reaching the resuspension of 2xPBS washed cells into dilution buffer, I processed the fecI samples (4 hr 30 min of growth with OD600 of 0.254 and 0.247).
I sheared all of the samples together, using 2x shearing for the 0.1% formaldehyde samples and 4x shearing for the 1% formaldehyde samples (as described in section 2.17.1 above).
After the shearing, I used quick DNA concentration method (5 ml proteinase K and 25 ml H2O at 65C for 1hr followed by Qiagen PCR purification), yields for the four samples were (eluted into 50 ml EB):
Sample DNA (ng/ul) 260/280 260/230
fecI 0.1% formaldehyde 125
fecI 1% formaldehyde 112
lrp 0.1% formaldehyde 136.3
lrp 1% formaldehyde 148.7
The mean yield was 131 ng/ml . I was a little surprised to see that the fecI yields were so close to the Lrp yields, since the Lrp OD600 was almost 2x the fecI value.
Also, I should mention that I only have 2 samples for each TF here, whereas my factorial experiment called for 8 samples. That's because I pulled two samples from each of the different formaldehyde amounts. I did this A) because I'm lazy and I could grow/shear half of the cultures and B) because this ensures that I start with the exact same sheared chromatin for the enrichment protocols, so there should be less noise when comparing the wash and bead factors on the same TF/formaldehyde combination.
Before going home, I started the overnite incubations with the dynal and agarose overnite samples with 2 mg and 10 mg of sheared chromatin respectively (as described in section 2.17.1 above). I used 2.75 ml (3.3 mg ) of antibody for each antibody sample. I left the other (40min:40min) samples in the fridge.
Thur Nov 8, 2007
I started the 2 hr bead incubations and then shortly thereafter set up the 40 minute antibody incubation followed by the 40 minute bead incubation. So both groups finished at the same time.

2.17.3  new lrp and fecI primer plates

Fri Nov 9, 2007
I wanted to make some better PCR primer plates for this fractional factorial experiments. In particular, I wanted to limit the qPCR error by using a qPCR technical replicate. To do this I moved from 11 random genes to 9 random genes (I just removed the last 3 genes from the list) and used 3 technical replicates of one target for each tf (fecABCDE for fecI and livK for lrp).
By switching to 12 primer pairs per TF, I also made it possible to run all 16 samples (8 lrp and 8 fecI) in a single 384-well plate.
The plates are:
Lrp techRep small primer plate
- 1 2 3 4 5 6
A gcl mog pinO idnD yhaF nhaA
B aimA goaG kdtB livK livK livK
FecI techRep small primer plate
- 1 2 3 4 5 6
A gcl mog pinO idnD yhaF nhaA
B aimA goaG kdtB fecABCDE fecABCDE fecABCDE
I used my typical concentration of 2 mM for each plate and made 400 ml of each primer pair in TE.

2.17.4  lrp and fecI fractional factorial results

Fri Nov 9, 2007
I ran the qPCR plate using the two new primer plates described in the previous section.
Click for little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see fecI_lrp_factorial.m for the script).
Below are the raw enrichment values for each of the tested factor combinations:
FecI and Lrp fractional factorial results
form (%) incubation time bead type high salt wash livK enrich fecABCDE enrich
0.1 40min:40min agarose no 1.0 1.5
1 40min:40min agarose yes 1.8 2.3
0.1 overnight:2hr agarose yes 2.0 1.9
1 overnight:2hr agarose no 1.0 1.1
0.1 40min:40min dynal yes 2.8 2.7
1 40min:40min dynal no 1.8 0.9
0.1 overnight:2hr dynal no -1.3 -0.5
1 overnight:2hr dynal yes 2.9 4.1
Brief Conclusions:   I haven't bothered to do a proper factorial analysis of these results, because the general idea just jumps right out of the table above (particularly after I added some italics to help them jump). The best performing methods are the same for BOTH TRANSCRIPTION FACTORS!!!!!!
The top two solutions are particularly interesting and thankfully they both use dynal beads (so the protocol can remain 96-well without coming up with a 96-well version for agarose beads). With a few extra experiments or perhaps just additional analysis of this data, I could tease apart what really mattered. However, I'm going to just use the top two solutions as-is, because they have nice properties.
The top performing combination is 1%:dynal:overnite:wash, which is a 2.5 day protocol.
The second best performing combination is 0.1%:dynal:40min:wash, which is a 1.5 day protocol.
One clear thing to take away, it that the high salt wash looks like it is certainly a crucial element to increase signal/noise (though I'm still curious why the LiCl based wash didn't do the job when I tested that by itself; my guess is that the 0.1% SDS in the high salt is really removing the background binding).

2.18  testing the 2.5 day, top-performing protocol, on Lrp, FecI, and Lrp

In the previous section, I was able to obtain maximal enrichment for FecI AND Lrp by using 1% formaldehyde, dynal beads, with an overnight antibody incubation, 2 hr bead incubation and a high salt wash. To determine if this was a fluke and to get an estimate of the error of this protocol, I'm going to try again (like I did in section 2.16) to run 6 replicates for each TF using this new protocol.
Sat Nov 10, 2007
I started an overnite culture of lrp, pdhR, and fecI
Sun Nov 11, 2007
I grew 3 x 50ml 1:100 dilution LB cultures in 250 ml baffled flasks for each transcription factor. Once again the fecI grew slower, so I processed the FecI samples after crosslinking and washing the Lrp and PdhR samples. Lrp and PdhR were grown 3 hr 20 min (0D600 0.77 0.751 0.767 for Lrp samples 1,2,6 and 0.726 0.642 0.68 for PdhR samples 5,7,8). FecI samples were grown 4 hr 40 min (OD600 0.46 0.506 0.399 for FecI samples 3,4,9).
excel file of the randomized experiment setups below
Randomized Growth setup
randID sample
1 lrp
2 lrp
3 fecI
4 fecI
5 pdhR
6 lrp
7 pdhR
8 pdhR
9 fecI
FecI, Lrp, PdhR six ChIP replicates
randID sample
1 fecI
2 fecI
3 pdhR
4 lrp
5 fecI
6 pdhR
7 lrp
8 lrp
9 pdhR
10 fecI
11 pdhR
12 pdhR
13 lrp
14 lrp
15 fecI
16 fecI
17 lrp
18 pdhR
One important thing to note. Upto this point, I'd always washed the cells 2x with PBS to remove trace amounts of PBS and Glycine before sonication in dilution buffer. However, I didn't have enough PBS or enough time to make more, so I washed 1x in 8 ml of PBS (normally I use 2x10ml). To maximize the amount of media/formaldehyde/glycine that I removed, I placed the centrifuge tubes upside down on a stack of paper towels for a minute or so..
I crosslinked with 1% formaldehyde, sheared 4x20%x30secs. I quantified the sheared chromatin by adding 5 ml proteinase K and 25 ml H2O to 100 ml of each sample and incubating at 65C followed by a Qiagen PCR purification. Average DNA amount was 97.87 ng/ml .
Raw nandrop data in excel file
Chromatin yields were (remember I used 100 ml EB here instead of the normal 50 ml ; it was by mistake, but I just have to remember to scale everything appropriately):
Sample ID ng/ul 260/280 260/230
1 63.67 1.79 1.73
2 73.82 1.79 1.8
3 69.24 1.75 1.88
4 63.9 1.8 1.73
5 71.97 1.73 1.71
6 80.36 1.82 1.79
7 132.77 1.85 1.78
8 79.25 1.82 1.66
9 99.12 1.82 1.57
10 109.41 1.9 1.69
11 123.87 1.86 1.66
12 126.3 1.87 1.71
13 127.44 1.86 1.66
14 139.56 1.86 1.68
15 86.77 1.89 1.47
16 77.02 1.83 1.49
17 120.02 1.81 1.6
18 117.13 1.86 1.7
I eluted into 100 ml EB. 20.5 ml of each sample (2 mg ) was mixed with 59.5 ml of dilution buffer and 2.75 ml (3.3 mg ) of Anti-Xpress antibody. The PCR strips were rotated overnite at 4C.
Mon Nov 19, 2007 I ran the gel (Figure ) after the qPCR and the analysis were done just to do a sanity check that the shearing lengths were fine (they were).
Please see the pdf version for figures
Figure 2.29: 1.5% agarose gel of sheared chromatin run for 37 minutes
Mon Nov 12, 2007
I added 100 ml of PBS/0.5%BSA washed beads and incubated for 2 hr at 4C. I washed the beads with 180 ml of high salt wash (5 min rotation at 4C). Finally I washed 2xTE at room temp and eluted into 180 ml of dynal elution buffer - leaving the samples at 65C for crosslink reversal.

2.18.1  more primer plates

Tues Nov 13, 2007
I'm going to make additional lrp, fecI, and pdhR primer plates to test these ChIP samples. I want to fit the PdhR and FecI samples on one plate, so I'm going to use the 11 random genes and one known target (aceE and fecABCDE for PdhR and FecI respectively). For lrp, I'm going to test most of the known targets I used to test my ChIP procedure for the PLoS 2007 paper in the first chapter of this notebook. To test the technical/qPCR noise relative to the sample noise, I included 3 PCR technical replicates of pntA, serA, livK, gltB. I also included one replicate of stpA.
PdhR primer plate
- 1 2 3 4 5 6
A gcl mog pinO idnD yhaF nhaA
B aimA goaG kdtB aceE citC fruK
FecI primer plate
- 1 2 3 4 5 6
A gcl mog pinO idnD yhaF nhaA
B aimA goaG kdtB fecABCDE citC fruK
Lrp primer plate
- 1 2 3 4 5 6
A gcl mog pinO idnD yhaF nhaA aimA goaG kdtB pntA citC fruK
B pntA serA livK gltB livK serA gltB gltB livK pntA pntA serA

2.18.2  2.5 day protocol results

Tues Nov 13, 2007
I ran the 6 FecI and 6 PdhR samples in the 384-well qPCR plate together.
I ran the 6 Lrp reactions on a second plate.
Please see the pdf version for figures
Figure 2.30: Boxplot of the old protocol vs the 2.5 day protocol.
Click for little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see fecI_pdhR_new_way2.m and lrp_new_way_2ndTry.m for the scripts).
Brief Conclusions:   It is done. The 2.5 day protocol after the final factorial optimization of FecI and Lrp is much faster, enrichs more, and is less noisy than the original protocol (Figure 2.30).

2.19  testing the 1.5 day, top-performing protocol, on Lrp, FecI, and Lrp

In section 2.17, I was able to obtain the 2nd best enrichment for FecI AND Lrp by using 0.1% formaldehyde, dynal beads, with a 40 min antibody incubation, 40 min bead incubation and a high salt wash. To determine if this was a fluke and to get an estimate of the error of this protocol, I'm going to run 4 replicates for each TF using this 1.5 day protocol.
Wed Nov 14, 2007
I started overnites of Lrp, FecI, and PdhR.
Thur Nov 15, 2007
I grew 50 ml cultures of all three TFs from dilutions of overnite in 250 ml baffled flasks. I used 1:100 dilutions for Lrp and PdhR. I used a 1:50 dilution for FecI (which grows slowly). I grew two of each TF. After 3 hr 10 min, the samples were taken with OD600: FecI (0.712, 0.673), Lrp (0.919, 0.896), and PdhR (0.825, 0.859) [I grew them a little longer than I would've liked].
I took two samples from each culture flask for 4 replicates total for each TF. The randomized order was:
lrp, lrp, fecI, pdhR, lrp, fecI, fecI, lrp, pdhR, pdhR, fecI, pdhR
I quantified the DNA by incubating 100 ml samples of sheared chromatin with 5 ml proteinase K and 25 ml H2O for 1 hr at 65C followed by a Qiagen PCR purification. The DNA was quite concentrated (presumably because I let the cultures grow to a higher than normal OD600). I used 16 ml (appx 2 mg ) with 64 ml dilution buffer for each immunoprecipitation. Based on the information from the nanodrop spec readings (raw spec data in excel):
Sample ID ng/ul 260/280 260/230
1 241.19 1.83 1.81
2 185.94 1.83 1.75
3 271.84 1.8 1.56
4 247.77 1.83 1.83
5 281.99 1.84 1.88
6 268.81 1.81 1.67
7 279.45 1.8 1.57
8 280.24 1.83 1.83
9 237.46 1.83 1.86
10 232.56 1.84 1.85
11 252.77 1.84 1.79
12 225.22 1.84 1.83
Fri Nov 16, 2007
I cleaned up the samples and ran the qPCRs all on one plate using the same primer plates as I used in the previous section.
Click for little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see old_new_compare_combine.m for the script). I analyzed the 1.5 day protocol and combined it with the original and the 2.5 day protocol (Figure ).
Fri Nov 19, 2007
I ran the gel (Figure ) after the qPCR and the analysis were done just to do a sanity check that the shearing lengths were fine (they were).
Please see the pdf version for figures
Figure 2.31: 1.5% agarose gel of sheared chromatin run for 30 minutes
Please see the pdf version for figures
Figure 2.32: Boxplot of the old protocol vs the 2.5 day vs the 1.5 day protocol.
Brief Conclusions:   The 1.5 day and 2.5 day protocols reflect the results obtained from the previous fractional factorial of Lrp and FecI quite well and both outperform the original protocol in general (Figure 2.32).

2.20  combinatorial epitope tags to improve consistency?

The optimizations thus far have exceeded my expectations. The factorial and response surface optimizations certainly helped move towards a much faster and better performing ChIP protocol. However one thing that I haven't address thus far is why some tagged transcriptions fail. The new protocols take the ChIP reactions that worked before and make them better, but how can we get the transcription factors that don't enrich with ChIP to work? In the past I've found on multiple attempts that LexA doesn't with ChIP (using both the original protocol and a partially optimized protocol about half-way through these factorial and RSM optimizations). I also have a tiny amount of effidence that PhoP and CysB don't work either (I tried them only one time). To determine the binding sites of all TFs, we're either going to need a gigantic custom built monoclonal library (my dream) we're going to need to figure out alternative for these tagged-TFs that don't enrich with ChIP.
My first guess as to why some TFs work and some TFs don't is that the tag is either inhibiting the binding of the TF to the genome or the TF protein folding is inhibiting the access of the antibody to the epitope tag. To test this idea, Ilaria (cloning master) Mogno, built four different combinations of epitope tags for three transcription factors (Figure ). The clones are on an extremely low-copy plasmid (3-5 copies/cell) with kan resistance.
Please see the pdf version for figures
Figure 2.33: Ilaria made tagged versions of lexA, fecI, and lrp using four different locations and combinations of the Xpress tag and the myc tag

2.20.1  testing the four epitope tag combinations with lexA, fecI, and lrp

Sun Jan 27, 2008
I started 3 ml of overnite for each of the 12 strains.
Mon Jan 28, 2008
I began the 1.5 day ChIP protocol with all 12 samples at around 11:30AM. I used a 1:100 dilution for the lexA and lrp strains and a 1:50 dilution for the fecI strains. After 1hr 30min, I added 10 mM of IPTG (this was a half-an-hour later than I meant to add it). After 2hr 35 minutes, the cells had reached the appropriate OD600 of around 0.5-0.8. The OD600 (not background subtracted) and sample order is in the table below.
Experimental design for first epitope tag tests
sampleID randomID sampleName OD 2hr 10min OD 2hr 35 min
7 1 XlrpN 0.439 0.62
8 2 XlexAN 0.425 0.617
2 3 XlexA 0.421 0.614
9 4 XfecIN 0.555 0.841
6 5 fecIN - 0.825
1 6 Xlrp - 0.638
4 7 lrpN - 0.641
5 8 lexAN - 0.544
10 9 lrpX - 0.648
11 10 lexAX - 0.62
3 11 XfecI - 0.839
12 12 fecIX - 0.851
For the strains with both an Xpress and a myc epitope, I used 2.75 ml (appx 3.3 mg ) of each antibody (so twice as much total antibody).
Brief Conclusions:   With the low copy plasmid, it appears that the fecI grows just as fast as the other two strains, so I should probably use 1:100 for all of the strains from now on.
Tues Jan 29, 2008
I cleaned up all of the 12 samples (24 total including the -antibody and +antibody).
I built a new primer plate that is essential the previous lrp primer plate with an additional row (C) of lexA targets in triplicate (for PCR technical replicates).
Lrp and LexA tag-test primer plate
- 1 2 3 4 5 6
A gcl mog pinO idnD yhaF nhaA aimA goaG kdtB pntA citC fruK
B pntA serA livK gltB livK serA gltB gltB livK pntA pntA serA
C sulA umuC dinF recAsulA umuC dinF recAsulA umuC dinF recA
Click for little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see lexA_lrp_factorial.m for the script).
Brief Conclusions:   Those preliminary qPCR results, hint that the lexA might be enriching for the first time! However, overall the qPCR reaction itself was pretty crappy (I was in a hurry and I don't think I filled the plate perfectly; quite a number of failed reactions).
repeating the qPCR reaction
Mon Feb 4, 2008
Given that so many of the PCR reactions failed in the previous qPCR plate, I decided to repeat the lexA lrp qPCR using the remaining ChIP DNA. I ran the exact same reaction, being extra careful no to make any mistakes while filling the plate.
Click for little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see lexA_lrp_factorial_rep2.m for the script).
Brief Conclusions:   qPCR reactions were much cleaner, and produced the same basic conclusions that the lexA might be working (and that the lrp works with the new tags as well). I'll add some figures and things after I get some replicates.
qPCR with fecI and the new tags
Tues Feb 5, 2008
I ran a qPCR of the fecI samples using the following fecI primer plate:
FecI tag-test primer plate
- 1 2 3 4 5 6 7
A gcl mog pinO idnD yhaF nhaA fecABCDE
B aimA goaG kdtB fecABCDE citC fruK fecABCDE
Click for little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see fecI_IM_tags.m for the script).
Brief Conclusions:   More positive results, seems the new tag library works work fecI as well. I'm not sure about the tagged transcription factors that contain only an Xpress epitope, but both of the tagged TFs that contain a myc tag worked (the tag with only a myc tag rather than both a myc and an Xpress tag worked best for this replicate). The version with a single myc on the N-terminal worked best for both lexA and fecI; the lexA version with N-terminal myc didn't seem to work and the values looked pretty odd, so we'll have to see how this looks in another replicate. These results are certainly promising enough to merit another replicate.

adding a second replicate of the four epitope tag combinations with lexA, fecI, and lrp

Fri Feb 8, 2008
I ran the 1.5 day ChIP protocol version 1.1 with the growing a 1:100 dilution of the same 12 strains in LB using the following randomized order:
sampleID randomID sampleName OD600
8 1 XlexAN 0.661
11 2 lexAX 0.612
9 3 XfecIN 0.613
12 4 fecIX 0.598
3 5 XfecI 0.633
1 6 Xlrp 0.675
6 7 fecIN 0.597
2 8 XlexA 0.555
10 9 lrpX 0.603
5 10 lexAN 0.577
7 11 XlrpN 0.605
4 12 lrpN 0.56
After 1 hr of growth 20 mM of IPTG was added to each sample (I meant to add 10 mM like last time, but I messed up and added 20 mM ). Samples were taken for crosslinking at 2 hr 30 min at an OD600 of around 0.6 (see table above; values not background subtracted).
Average sheared chromatin was around 100 ng/ml , so I used 20 ml of sheared chromatin and 60 ml of dilution buffer for the immunoprecipitations.
Sat Feb 9, 2008
I cleaned up the 24 rxs and ran the qPCR plate for the lrp and lexA strains.
Sun Feb 10, 2008
I ran the qPCR plate for the fecI strains
Click for little info about how I analyzed the above data and the exact matlab batch scripts I used are in the same directory (see lexA_lrp_factorial_completeRep2.m and fecI_IM_tags_completeRep2.m for the scripts).

combining the combinatorial epitope tag results

I'm going to combine the two factorial results into a single table with the median value for each gene for each replicate. For replicate 1, I'm going to use the second qPCR rxn for lexA and lrp where far fewer rxns failed.
transcription factor target log enrich rep1 log enrich rep2
XlexA sulA -0.036 0.023
lexAX sulA 0.027 0.014
lexAN sulA 0.1801 0.985
XlexAN sulA 0.328 0.533
XlexA umuC 0.032 0.059
lexAX umuC -0.056 0.008
lexAN umuC -0.355 0.961
XlexAN umuC 0.100 -0.101
XlexA dinF 0.108 -0.071
lexAX dinF -0.032 0.084
lexAN dinF 1.5172 0.879
XlexAN dinF 0.867 0.650
XlexA recA -0.012 -0.086
lexAX recA 0.006 -0.014
lexAN recA 1.570 0.497
XlexAN recA 0.143 0.343
fecIX fecABCDE 0.080 0.011
XfecI fecABCDE 0.024 0.078
fecIN fecABCDE 1.722 2.659
XfecIN fecABCDE 0.639 1.599
Xlrp pntA -0.085 -0.022
lrpX pntA 0.086 -0.232
lrpN pntA 0.362 0.105
XlrpN pntA 0.168 0.120
Xlrp serA 0.420-0.083
lrpX serA 0.4620.233
lrpN serA -1.989 0.324
XlrpN serA 1.032 1.060
Xlrp livK -0.742 0.046
lrpX livK 0.789 0.343
lrpN livK -2.001 1.287
XlrpN livK 2.485 2.544
Xlrp gltB 1.001-0.036
lrpX gltB 0.301 0.044
lrpN gltB -4.666 1.000
XlrpN gltB 0.922 1.265
I think the table makes more sense if sorted by tag:
transcription factor target log enrich rep1 log enrich rep2
XlexA sulA -0.036 0.023
XlexA umuC 0.032 0.059
XlexA dinF 0.108 -0.071
XlexA recA -0.012 -0.086
XfecI fecABCDE 0.024 0.078
Xlrp pntA -0.085 -0.022
Xlrp serA 0.420-0.083
Xlrp livK -0.742 0.046
Xlrp gltB 1.001-0.036
lexAX sulA 0.027 0.014
lexAX umuC -0.056 0.008
lexAX dinF -0.032 0.084
lexAX recA 0.006 -0.014
fecIX fecABCDE 0.080 0.011
lrpX pntA 0.086 -0.232
lrpX serA 0.4620.233
lrpX livK 0.789 0.343
lrpX gltB 0.301 0.044
lexAN sulA 0.1801 0.985
lexAN umuC -0.355 0.961
lexAN dinF 1.5172 0.879
lexAN recA 1.570 0.497
fecIN fecABCDE 1.722 2.659
lrpN pntA 0.362 0.105
lrpN serA -1.989 0.324
lrpN livK -2.001 1.287
lrpN gltB -4.666 1.000
XlexAN sulA 0.328 0.533
XlexAN umuC 0.100 -0.101
XlexAN dinF 0.867 0.650
XlexAN recA 0.143 0.343
XfecIN fecABCDE 0.639 1.599
XlrpN pntA 0.168 0.120
XlrpN serA 1.032 1.060
XlrpN livK 2.485 2.544
XlrpN gltB 0.922 1.265
Brief Conclusions:   It's pretty noisy, but clearly the XtfN and tfN work the best. I think I want to continue on with both versions and run some replicates using the 2.5 day protocol. With the current data is seems that fecI definitely works better with a N-terminal myc only; lrp works better with the the C-terminal Xpress and the N-terminal myc; and lexA I can't tell. In general it looks like the second rep worked better than the first perhaps because of the added IPTG? Perhaps I should beef it up even further with the 2.5 day samples (perhaps 50 mM or 100 mM ).

2.21  END OF COMPLETED SECTIONS

Note that all sections below this one are ideas I've either not completed or I've decided not to complete.

2.22  Cool antibody factorial experiment to maybe try someday

2.23  Cloning lrpB into a myc-tagged plasmid

Thur Jun 21, 2007
I want to try lrp with the tag on the other end. I'm also going to use the myc tag instead of the Xpress tag. I ran the following PCR reaction for cloning in the pTrcHis from invitrogen: Easy-A master mix 10 ml , 7.5 ml H2O , 2 ml primers (from 10 mM stock), 0.5 ml lrpB plasmid. Note that the lrp-myc primers remove the last codon of the gene so that the myc tag can be translated.
I used 1 ml of the PCR in the TOPO rxn, and I cloned the plasmid according to Invitrogen's instructions.

2.23.1  miniprepping and sequencing the new lrp plasmids

I grew up 3 lrp low-copy and 4 lrp-myc plasmids and miniprepped them. I eluted into 50 ml Yields were (lc = low-copy, m = myc):
Sample DNA (ng/ul) 260/280 260/230
lc-A 69.2
lc-B 59.9
lc-C 93.5
m-A 150.3
m-B 100.2
m-C 111.7
m-D 104.5
I cut 10 ml of the miniprepped DNA to check the insert size (Figure ). All of the insert sizes looked appropriate.
Please see the pdf version for figures
Figure 2.34: gel of cut lrp lowcopy and cut lrp myc

2.24  NOTES

matrix to test formaldehyde vs shearing:
form: 0.1, 1, 2, 4
shear: 2, 3, 4, 5

2.25  list of annoying things I'd like to check with a final factorial

  1. do I really need to wash the beads with BSA? Can I just use PBS? Or citrine?
  2. do I really need the two TE washes? or can I just add elution buffer and be done?
  3. do a incubation time experiment where I check 5 min, 10 min, 40 min, and 80 min for a final sanity check if incubation time matters
  4. can I do the bead/antibody incubation at RT?
  5. do I really need to wash the cells 2x with ice cold PBS after formaldehyde and quenching? is one time enough? is the wash necessary at all? why not just add dilution buffer?
  6. do I really need to place the dilution buffered beads at 65C (annoying with PCR strips)?
  7. is proteinase K necessary during the crosslink reversals? (try both with 100 ml of sample; run a paired-ttest with the samples)
  8. try the low pH buffer recommended by dynal
  9. try the His Antibody + Xpress?
try: BSA beads (y/n), low pH beads (y/n), TE (2x/0x), wash with PBS (2x/1x), proteinase K (y/n), low pH dilution (y/n), elution at (65C/RT), dilution buffer (TE + RNAse/standard with NaCl and Triton X100 + RNAse)

2.25.1  other things to check

does a lower pH matter? should I resuspend the beads in the lower pH. would be nice to use that 96-well PCR strip magnet that also allows in place mixing