Genomic Research Laboratory
Group of Prof. J. Schrenzel
Service of Infectious Diseases
Geneva University Hospitals (HUG)
CH-1211 Geneva 4, Switzerland

Home Research Microarray Software Publications Media Teaching Contacts


ZeptoCHIP: A Microarray for Rapid Bacterial Identification from Blood Culture bottles

The ZeptoCHIP is an oligonucleotide microarray designed to detect and identify more than 97.5% of bacteria responsible of bacteraemic episodes. This microarray, profiting from the Planar Waveguide technology for increased signal/noise performances, was developed through a close collaboration between ZeptoSens AG (Witterswil - Switzerland) and our laboratory.

This study led to the development of:

  1. a bioinformatics strategy to select short oligonucleotide signature probes for bacterial identification;
  2. a rapid procedure for target labeling and hybridization to the array;
  3. a customized artificial neural network and decision tree for microarray data analysis.


Bacterial identification from blood cultures requires sequential steps:

  1. detection and sub-plating of blood bottles that display metabolic activity,
  2. growth of pure culture(s), and
  3. bacterial identification using appropriate phenotypic tests.

This strategy relies on bacterial growth and requires at least 24-48h.
Earlier results could rationalize the usage of broad-spectrum antibiotics, thereby reducing drug-related selection pressure and costs.
The goal of our study was to develop a rapid bacterial identification directly from blood culture samples (less than 6 hours).


Analysis of positive blood cultures identified in our institution during the last decade Selection of a subset of species covering >95% of bacteriemic episodes.
Define unique molecular signatures for these pathogens based on their 16s rRNA.
Assess the discrimination of these signatures on microarrays.
Develop a highly sensitive detection method to skip enzymatic amplification.
Compare conventional analytical methods with artificial neural networks.

Microarrays design

Fig. 1 : 16s rDNA sequence database.

Figure 2:
Top window:
user-defined input parameters: calculated Tm; calculated probe sequence composition (%GC); weighting coefficient for mismatched positions; number of signatures per target. Bottom window: information on selected probes for : GC content (A) and calculated Tm (B). Left panels: probes selected in a subset of 350 genes. Right panels: probes selected as a percentage of all genes. (C) Shape of the weighting curve: the 5’-end of the probe is depicted on the left of the plot.
[Click image to enlarge]
Click to enlarge
Fig. 2 : Selection of discrimant signatures.

Click to enlarge
Fig. 3 : ZeptoCHIP scanning (Klebsiella oxytoca and Klebsiella pneumoniae )

Images obtained from the scanning of six separate ZeptoCHIP experiments, either using Klebsiella oxytoca (top row) or Klebsiella pneumoniae (bottom row) as a starting material. The ZeptoCHIP arrays contained 272 spotted features. Six probes provided reproducible signals in both Klebsiella spp (yellow circles) while four other probes (orange circles) provided additional information that allowed distinguishing both species.
[Click image to enlarge]

Cluster analysis

Click to enlarge Click to enlarge
One-way clustering of 63 experiments, using 8 bacterial classes as different sources of total RNA. Features were background subtracted, and clustered using a Pearson correlation in GeneSpring (Silicon Genetics).
Color-coded vertical bars correspond to the signals generated by each of the 272 probes.
[Click image to enlarge]
Synthetic display of the same experiments, but here the clustering was performed after data normalization. Each class is depicted by a separate color.
[Click image to enlarge]

Decision tree

Click to enlarge Decision tree generated by CART, using the same data as in the clustering analysis (same color code). Each node corresponds to a branching point that categorizes experiments. Branching stopped when the classification improvement reached an optimal cost.
[Click image to enlarge]

Neural network

Estimated probabilities for the experiments to be misclassified by an artificial neural network, after 600 learning cycles. For each bacterial species, percentages were determined using either a complete set of 272 probes (top bars), or a reduced set of 81 probes designed to detect only these 8 bacterial classes (bottom bars).
[Click image to enlarge]
Click to enlarge


Our strategy permitted to:

  1. Identify discriminant signatures in the 16s rRNA target,
  2. Select signatures with homogeneous thermodynamic properties.

We have established protocols for:
  1. Reliable coupling chemistry between probes and microarray,
  2. Robust chemical labeling and short hybridizations (<1h).

The ZeptoCHIP, an oligoarray with 272 probes designed to detect >95% of clinical pathogens, provided poor discrimination when using threshold-based analysis (clustering, decision trees) Artificial neural network analysis proved far more reliable for bacterial identification and discrimination. In collaboration with EPFL partners (Horst Vogel, Theo Lasser) and ZeptoSens AG, evanescent waveguide technology coupled to innovative signal amplification is under development to detect a few copies of hybridized target.


Rapid bacterial identification using evanescent-waveguide oligonucleotide microarray classification.
Francois P, Charbonnier Y, Jacquet J, Utinger D, Bento M, Lew D, Kresbach GM, Ehrat M, Schlegel W, Schrenzel J.
J Microbiol Methods. 2006 Jun;65(3):390-403 [Abstract] - [PubMed]

Research Projects
  qMRSA: Detection of MRSA
  MagRSA: Automated Diagnosis of MRSA
  S. aureus biofilms
  S. aureus Intracellular Survival
  S. aureus Proteomics
  NRP-49 Projects: Antimicrobial Resistance
  GESNOMA: Noma disease
  MLVA: S. aureus Genotyping
  MIF Knock-Out Mouse Macrophages
  Patho-adaptation of S. aureus
  Phylogenetic Microarray
  High Throughput Sequencing

Revised Feb 12, 2007 - © 2003-2023 Genomic Research Laboratory, Geneva