Ken Hyer, U'S' Geological Survey

1
Bacterial Source Tracking Methods Comparison and
Field Application
Ken Hyer, U.S. Geological Survey Richmond, VA
2
VA Dept of Conservation and Recreation
VA Dept of Conservation and Recreation
WV Department of Environmental Protection
WV Department of Environmental Protection
WV Department of Agriculture
WV Department of Agriculture
Fairfax County, VA
3
Objectives for Talk

• Describe a methods comparison study that
  evaluated 7 source tracking methods.
• Describe a field application of BST and the
  associated quality control activities.

4
Bacteria Source Tracking
5
BST Methods Comparison Study

• Sampling in Berkeley County, West Virginia.
• Involves researchers from across the nation and 3
  different USGS district offices (Melvin Mathes of
  WV and Don Stoeckel of Ohio).
• Five Genotypic Methods (and investigators)
• Ribotyping using two different enzyme sets
• (George Lukasik, Mansour Samadpour)
• Pulsed-field Gel Electrophoresis
• (West Virginia Department of Agriculture)
• rep-PCR using two different primer sets
• (Howard Kator, Don Stoeckel)
• Two Phenotypic Methods (and investigators)
• Antibiotic Resistance Analysis (Bruce Wiggins)
• Carbon Substrate Utilization (Chuck Hagedorn)

6
BST Methods Comparison Study

• Prominent sources of fecal pollution being
  considered (based on NRCS input for Berkeley
  County)
• Humans
• Cattle (beef and dairy)
• Chickens
• Swine
• Horses
• Dogs
• Canada Geese
• Deer

7
Methods Comparison - Study Design

• Collect feces from at least 20 individuals per
  source.
• Isolate and confirm a library of known E. coli
  from the fecal samples
• Total of 70-100 confirmed E. coli per source
• Total known library size of 900 isolates
• Prepare a blind sample set comprised of 200
  isolates that included three subsets
• 26 Replicates from the known library
• 150 Fresh isolates from the 9 prominent sources
• 24 Fresh isolates from sources that were not in
  the original known library (mice, cats, raccoons,
  etc.)

8
Methods Comparison - Study Design

• Each researcher identified the source of each
  blind isolate.
• Results were scored and the following were
  considered for each method
• Accuracy of isolate identification
• Precision (reproducibility of replicate isolate
  analyses)
• Robustness (isolates from sources not in the
  library are identified as unknown)

9
Methods Comparison - Results

• In a general sense, we found that
• -In this study, under these conditions
• -Most methods did not perform as well as we
  expected, based on published literature.
• -Detailed study manuscript is in press at
  Environmental Science and Technology.

10
Results - Replicates

• The first 3 methods used discriminant analysis
  (DA), the other 4 used direct matching
  techniques.
• In scoring the replicates, the response unknown
  was considered incorrect (all isolates were in
  the known library).
• For each method, considering an 8-way source
  classification
• -ARA 6 of 26 (23 correct)
• -CUP 6 of 25 (24 correct)
• -RT-HindIII 3 of 23 (13 correct)
• -RT-EcoR1 14 of 26 (54 correct)
• -PFGE 24 of 24 (100 correct)
• -BOX-PCR 17 of 26 (65 correct)
• -REP-PCR 10 of 23 (43 correct)

11
Results - Accuracy

• For the accuracy subset, the response unknown
  was considered neutral (neither correct nor
  incorrect) thus the number of isolates
  attempted is very important.
• For each method, considering an 8-way source
  classification
• -ARA 36 of 150 (24 correct)
• -CUP 20 of 143 (14 correct)
• -RT-HindIII 19 of 147 (13 correct)
• -RT-EcoR1 7 of 8 (88 correct)
• -PFGE 15 of 40 (38 correct)
• -BOX-PCR 32 of 149 (21 correct)
• -REP-PCR 23 of 93 (25 correct)

12
Results - Ringers

• In scoring the ringers, the response unknown
  was considered the only correct response because
  none of these isolates were in the library.
• None of the DA methods attempted to identify and
  reject ringers.
• For each method, considering an 8-way source
  classification
• -ARA 0 of 24 (0 correct)
• -CUP 0 of 24 (0 correct)
• -RT-HindIII 0 of 24 (0 correct)
• -RT-EcoR1 24 of 24 (100 correct)
• -PFGE 16 of 24 (67 correct)
• -BOX-PCR 0 of 24 (0 correct)
• -REP-PCR 8 of 24 (33 correct)

13
Reasons For Method Underperformance

• Inadequate library size or structure
• Temporal component in the source-library
  collection
• Presence of many repetitive subtypes (transient
  strains)
• Different statistical analyses may be needed
• Regardless of these possible reasons, the study
  clearly demonstrates the need for QA/QC and
  proofing of methods.

14
Methods Comparison - Conclusions

• This is only one of several ongoing comparison
  studies. It demonstrates that under these study
  conditions, none of these methods are performing
  at the levels we anticipated.
• We can offer these recommendations
• -Perform considerable QA/QC in your BST work!
  This may include (1) analyzing blind collections
  of known isolates, (2) use of multiple BST
  methods, and (3) the use of other tracers to
  support the BST work.
• -Perform your QA/QC in such a way that you can
  detect if your method is working or failing.

15
Example of an Applied Study
16
Study Design

• Field Data Collection
• Water-sample collection
• Baseflow
• Stormflow
• Continuum
• Source sample collection
• Bacteria Source Tracking Analysis (Ribotyping)

17
Results of MSTBy Individual Contributor
18
Seasonal Patterns in MST DataComparison of Warm
and Cool Seasons
19
Validation of MSTHuman Signature
20
Validation of MSTPoultry Signature
21
Comparison of MST ResultsComparison of Accotink
Creek and Four Mile Run
Accotink Creek, BST Results
Four Mile Run, BST Results
(N279)
(N278)
22
Take Home Messages

• Perform considerable QA/QC to ensure that you
  have confidence in your results.
• Many different tools that can be applied to
  quality assure your BST data.
• Under appropriate conditions, it appears BST can
  be used to successfully identify bacterial
  sources.

23
USGS Contact Information
Ken Hyer 1730 E. Parham Rd Richmond, VA
23228 Email kenhyer_at_usgs.gov Phone
804-261-2636 On the web http//va.water.usgs.gov/