Skip Navigation

Principal Investigators Meetings

Vertical Array Project
Rosana Risques, Martin Judex, Gaelle Rhondeau, John Welsh, Michael McClelland, Charles Berry, Thomas Werner, Miguel Peinado
Table of Contents:

Vertical Array Project
Sidney Kimmel Cancer Center, San Diego

Rosana Risques SKCC
Martin Judex SKCC
Gaelle Rhondeau SKCC
John Welsh SKCC
Michael McClelland SKCC
Charles Berry UCSD
Thomas Werner Genomatix
Miguel Peinado Barcelona



Concept

Coverage of thousands of experimental variables using microarrays is expensive and requires thousands of array hybridizations.

QPCR is great for thousands of variables, but can become costly when hundreds of genes must be examined.

"Vertical arrays" are modified dot blots that are economical when thousands of variables and hundreds of genes are to be examined.

The choice between standard “horizontal” arrays, QPCR, and vertical arrays is a matter of cost benefit analysis




Vertical Array Strategy

  1. Make low complexity representations of the mRNA populations.
  2. Spot these on a glass slide.
  3. Probe one gene at a time, or a few using multichannel fluorescence.



RNA Arbitrarily Primed PCR RAP-PCR to make LCRs

RNA Arbitrarily Primed PCR RAP-PCR to make LCRs



RAP-PCR is robust

RAP-PCR is robust

Triplicates of six time points, serum-starvation and re-feeding.

The fingerprint shows 50-100 bands.

Underlying the visible bands are about a thousand bands that are too faint to see.

These faint bands come preferentially from the complex class and are reproducible.




Rot curve

Rot curve

These are at a disadvantage because their low sequence complexity results in only a few poor matches.




RAP + Array

RAP + Array

The RAP-PCR reactions, themselves, are analyzed by hybridizing to a cDNA array.




Different RAP probes (red) vs. oligo dT probe (green)

Different RAP probes (red) vs. oligo dT probe (green)



Oligo dT vs RAP

Oligo dT vs RAP



RAP vs RAP

RAP vs RAP



RAP vs. QPCR

RAP vs. QPCR

Genes selected from serum-starvation re-feeding experiment.




Cumulative Coverage of 3800 Genes

Cumulative Coverage of 3800 Genes

(5 best re-sampling average = 1.7)




Comparison with Affymetrix GeneChip

  Stdev above background
  3 10 100

Genes in common 530 471 378
Present in GeneChip 312 291 231

  218 180 147
% missed by GeneChip 41 38 39
Corrected for library errors* 20 19 19
 
*The procedure enriches for library errors.

RT-PCR across splice junctions confirmed presence of 7 out of 10 messages detected by RAP but undetected by Affy. 3 remaining have not yet been sequenced-confirmed.




RAP-Array Cassette

(for multiple RAP hybridizations)
RAP-Array Cassette



Vertical Array Design

Each experiment is represented by ~10-20 spots made using RAP-PCR.

Vertical Array Design




Vertical Array Throughput

Each gene is represented in at least one spot per experiment

Three genes at a time can be studied in thousands of experimental contexts.

Vertical Array Throughput



Actin vs. Thrombospondin

Actin vs. Thrombospondin

Fibroblasts from 23 patients and 20 primers at 2 RNA concentrations + controls.




Actin vs. Cyclophilin

Actin vs. Cyclophilin

(different actin probe)




Actin vs. Cyclophilin Ratios

Actin vs. Cyclophilin Ratios



Technical notes

  1. Two-color comparison cannot be implemented in the usual way. Multiple randomly selected genes will be used to calibrate the amount of hybridizable DNA in a spot. Possibly in pools.
  2. Redundancy in data occurs because LCRs overlap.
  3. Number of RAPs per condition depends on required coverage of exemplar set. Estimate ~20 for 98%.



Uses of Vertical Arrays

  • Transcription Factor Predictor: Classification of exemplars, and co-classification of novel cases using vertical arrays to survey thousands of pleiotropic agents or cell lines
  • Drug screening: Survey of the impact of thousands of drugs on hundreds of genes.
  • Large sample screening: Survey hundreds of genes in thousands of pathology samples.
  • Experiment Archive: Keep representation of mRNA populations from many diverse experiments and laboratories for unplanned discovery as insights evolve.



Transcription factor prediction
(R33)
  1. Use transcription factor target ‘exemplar’ genes and experimental perturbation using drugs and/or different cell lines to build a model of exemplar gene behavior.
  2. Select perturbations or cell lines that best distinguish the exemplars from non-exemplars.
  3. Use these conditions in standard microarrays to discover targets for the transcription factor.



Exemplars

Exemplars



Drugs

Table 3: Agents Reported to Alter AP-1 or NF-kappaB controlled genes
Agent AP1 Effect AP1 Ref.* NF-kB Effect NF-kB Ref.** Agent AP1 Effect AP1 Ref.* NF-kB Effect NF-kB Ref.**
calphostin - 1 - 1 ionomycin + 12 ± 10
calyculin     + 2 lavendustin - 1    
curcumin - 2 - 3 lipopolysaccharide + 13 + 11
cyclic AMP + 2     lovastatin - 14    
cycloheximide     + 4 mitoxantrone     + 12
dexamethasone ± 3,4 - 5 N-acetylcysteine - 15 - 13
dithiocarbamate + 22 - 23 neopterin     + 14
dihydrolioate     - 2 okadaic acid     + 15
forskolin + 5     pertusis toxin - 16 - 16
genistine - 1     PMA + 18 + 11
GM-CSF + 6     pyrrolidine - 15 - 17
H2O2 - 7 + 6 staurosporine - 19 ± 18,19
herbimycin - 8 - 7 tepoxalin     - 20
hymenialdisine     - 8 thapsigargin + 20    
IGF I + 9     TNF-alpha + 21 + 21
IL-1 + 10 + 8 tyloxapol     - 22
*See References for AP-1 agents in Table 3 in Literature Cited
**See References for NF-kappaB agents in Table 3 in Literature Cited





Experiment size

(for single cell line)

Pairwise combinations: 33!/2!31! + 33 = 561
Time points: 4
RAPs per experiment: 10
Control spots: 60
Total spots: 22,470
   
Hybridizations: 108 genes + 20 control genes
   
Standard array equiv: 2244 hybridizations
Q-PCR equivalency: 242,352 reactions

Biomek FX robot to set up reactions, GeneMachines robot for spotting.