ISS Seed Screening Discussion

Our ISS seed screening process is powered by food safety math, algorithms and analysis. Here, we offer some additional discussion of why our safety process for sprouting seeds works so well.

Ground Rules

  1. When it comes to seed sampling for human pathogens, distribution, sample size, and total contamination per lot are the factors that determine the probability of capturing a pathogen.  Actual probabilities of detection require a knowledge of the quantity and distribution of pathogens in the particular seed lot. This information is not known.
  2. There’s a possibility that two or more pathogens could be lodged on one seed.  There is also a remote possibility that some contaminated seed could be clumped together.  So we base probabilities on contaminated seed units (CS), which include clumps, rather than CFU. Seed, while being processed, does get mixed well enough for very accurate seed sampling estimates.
  3. For ease of explanation, throughout the rest of this discussion all samples will be 25 grams and all bags will be 25 kilogram bags.  So the sample size is always 1/1000th of a bag or one seed per thousand seeds.
  4. Although random sampling is generally the accepted practice in sampling, when sampling for the presence/absence of pathogens in seed, it is better to sample every bag.  Although bags containing any pathogens may not be homogeneously distributed throughout a seed lot, the seed within each bag is fluid and usually very well distributed in the bag.

Larger Numbers of Samples Increase the Probability of Capture at a Given Contamination Rate.

Considering that we sample at least 1/1000 of the seed, the larger the lot, the larger the sample.

If seed is contaminated at 1 seed per kilogram, pulling one sample from one bag would only give you a 2.5% chance of finding it.  But if you didn’t find it in the first bag, you have another change, with identical odds in the second bag.  And the more bags there are, the more 2.5% chances you get at capturing a pathogen.  You get a 2.5% chance enough times, say 400 times, and you have a 99.9% chance of capturing at least one of those contaminated seeds.

ContaminatedBags &Kg ofContaminatedProbability
SeedsSamplesSeedSeeds perof Capture
per kg(Same #)SampledLot
110.025252.5%
1100.2525022%
14011,00063%
11002.52,50092%
14001010,000>99%

Table 5. Different size lots in which all lots are contaminated a the rate of one contaminated seed per kilogram.  The probability of capturing a pathogen increase as the contaminated seed per lot and the amount of seed sampled increase.

OK, What if the Seed Isn’t Evenly Distributed and Pathogens are in Just a Few Bags?

Uneven distribution is a problem if you are trying to find a particular level of contamination, but is not as much of a problem if you are trying to find any at all.  What matters most is the total number of contaminated seeds in the lot, not how many bags the contaminated seeds are in.

The chart below shows what happens if you sample various size lots, in which all lots have 2000 contaminated seeds, evenly distributed.  Notice that if you have one contaminated bag you have 2000 contaminated seeds, and one pull will capture one or more of those seeds 86.5% of the time if distribution is even within that one bag.  If you divided those 2000 contaminated seeds among 2 bags, you would have 1000 seeds per bag, but you have doubled the number of pulls, and the odds even out.

BagsBagsContaminatedKGOdds of
in LotSampledSeed per BagSampledDetection
1120000.02586.5%
2210000.0586.5%
10102000.2586.5%
100100202.586.5%
8008002.52086.5%

Table 6. Even Distribution among five different lots with the same

(2000) total number of ContaminatedSeeds (CS/lot, not CS/kg or CS/bag)

Our charts are actually based on the odds of capturing cleanseed.  Then we reverse the numbers in order to predict the number of contaminated seed.  If you try this with small numbers, such as a few marbles in a cup, the odds will change as you move the marbles from cup to cup.  However, there are about 12 ? million alfalfa seeds in a bag.  In this example, each bag in the lot that contains 800 bags has only 1997.5 more clean seeds than the lot with just one bag.  That is a difference of only 0.00016%.

So the number of clean seeds remains virtually unchanged, and the probabilities, for all practical purposes, remain the same.

This next chart shows a large lot which, except in the bottom row, is unevenly distributed among 800 bags but does assume even distribution in the bags that do contain contaminated seeds.  The first row has all 2,000 contaminated seeds in one bag.  The odds, as in the previous chart are 86.5%.  But it does not matter that you pulled seed from the 799 bags that are clean.  You pulled one pull, from the one bag, that was so contaminated, that it gave you an 86.5% chance of finding one contaminated seed.

Number of
ContaminatedSamples
BagsSeeds EachTakenOdds
NotBagsContaminatedin theof
ContaminatedContaminatedBagLotDetection
7991200080086.5%
7982100080086.5%
7901020080086.5%
7001002080086.5%
08002.580086.5%

Table 7. Uneven Distribution in an 800 bag lot. If only a few bags are contaminated, they need to be verycontaminated in order to have a good probability of capturing a pathogen.  If many bags are contaminated, likelihood of finding a contaminated seed is strong even if contamination is low.

It should be noted that if you take those 1, 2, 10, 100, or 800 contaminated bags and put differentlevels of contamination in each bag you will still have an 86.50% chance of capturing a pathogen as long as there are 2000 contaminated seeds in the lot.

The assumption below is that the seed lot has 20,000 kg in 800 25kg bags from which 25 grams is sampled from each bag.  There are 440,000 seeds per kg, or 8,800,000,000 (T) seeds per lot.  1,000th of the seed is pulled, giving a composite sample (N) of 8,800,000 seeds.  As you can see, there needs to be about 5,000 contaminated seeds in the lot in order to get a 99% probability of capture.  How many contaminated seeds are in any given lot that is contaminated is anybody’s guess.  There is no technology available to give even a good guess much less an accurate figure.  Also, this is a theoretical model that assumes even distribution, not throughout the lot, just throughout the contaminated bags.

Probability (P) = 1-(C/T)^N, where (C/T) = assumed ratio of Clean seeds to Total, and N = Number of seeds sampled.

Contaminated Seeds per LotProbability of Capture
                  101%
                  202%
                 20018%
              1,00063%
              2,00086%
              5,00099%
            10,000>99%

Table 8.

Probability of Capturing a Contaminated Seed in a Twenty-Ton Lot of Alfalfa Seed Contaminated with Various Contamination Levels Per Seed Lot.

Some Question the Reliability of Seed Sampling Because There is a Possibility that the Contaminated Seeds Could be in a Corner of One Bag or in a Clump and Never be Detected.

To be clear, although these probabilities are not dependant on even distribution throughout the lot, the probabilities are dependent on even distribution throughout the bags that are contaminated.  So if all 10,000 contaminated seeds in Table 8 are in the corner of one bag, the likelihood of capturing one or more of the contaminated seeds is very slim.  If it is in the corner 100 bags the odds are still small.  But if there is contamination in more than a bag or two, and, if it is a corner of the bag, it would most likely have occurred after the seed was bagged (mice eating their way through the bag).  That is why bag inspection is an important part of seed screening.

We have to assume that these things happen from time to time and there is not always even distribution in the contaminated bag or bags.  In this case, the probability charts are useless.  But this scenario may not occur as often as one might think. The seed is harvested, transported, and dumped into a silo or bins.

sprouting_seed_production

Figure 1.  Seed is mixed during harvest.

It is then poured or augured into the seed cleaning equipment, processed, and poured into a bag.  The cleaning and grading process does not allow even two seeds to clump together or they won’t fit through the screens.  Seed with pathogens are not likely to stay next to each other throughout this process.  They will be somewhat, if not thoroughly distributed.

Figure 2.  Seed is mixed during processing.

When Trying to Detect PlantPathogens, Sample Sizes are Used and the Probabilities are Extremely High as Well.

When looking for plant pathogens, you are looking for frequency rather than any at all. In order to determine the percentage of pathogens in wheat, 8 kg is sampled for each 100 tons of seed.  Then 300 seeds are pulled from that 8 kg for testing.  Distribution of the plant pathogens is good enough that this method is very accurate.  In the same lot size, we would sample and inspect the lot, and then 25 million seeds would be tested in order to find a single human pathogen!

Does Blending Make ISS Seed Sampling Less Reliable?

Yes and no.  In the following example, two bags of seed, contaminated at the rate of 4 Contaminated Seeds/kg contain 200 contaminated seeds.  If these two bags are blended in with 800 bags of non-contaminated seed, the new lot still has 200 contaminated seeds.  The probability of detection, by pulling 802 samples is 18.5%.  These are the same odds as if you had pulled one sample from each of the two contaminated bags before they were blended in.

Large LotSmall LotCombined
Bags8002802
CS/kg040.01
CS/lot0200200
Samples8002802
Seed Sampled1/10001/10001/1000
ProbabilityN/A18.5%18.5%

Table 9. If a contaminated and non-lot is combined, the odds of capture will remain the same if the percentage (1/1000 in this example) remains the same.

However, using the ISS Seed Screening Protocol on the two bags, 3 kg would have been pulled for testing (instead of just two 25 gm samples).  This increases your sample size from 1/1000 to 1/17th.  The odds of capture go to 99.99%.

 

Small Lot

Bags

2

CS/kg

4

CS/lot

200

Samples   (3 kg Total)

120

Seed Sampled

1/17

Probability

99.999%

Table 10. Increasing the sample size from 2 pulls to 120 pulls, taken from the 2 contaminated bags prior to blending, substantially increases the probability of capturing a contaminated seed.

So it does not make any difference if the lot is blended or not, as long as the protocol is followed prior to blending.  Or, if there is more than 120 bags (25gm x 120 = 3kg) in each lot that made up the blended lot.

Why is it Hit and Miss When Health Officials Try to Find Contamination in Seed That They Are Certain Caused an Outbreak?

Sample size, distribution, and the total number of contaminated seeds in the lot determine probability of capture.  Seed companies can screen the seed when the entire lot is in tact.  This is when the sample size will be greatest and the total number of contaminated seeds is at its highest.

By the time the epidemiologists determine that sprouts are most probably the cause of an outbreak, a good portion of that lot is gone, and a portion, if not all of the contaminated seed was used up in the outbreak.  The outbreak could have easily consumed the entire number of contaminated seeds needed to have a high probability of capture.

What Happens to the Odds When the Minimum Sample Size Requirement is in Play?

The protocol requires drawing at least 3 kg for testing.  But let’s suppose it didn’t.  If there is light contamination, say 4 contaminated seeds per kilogram, and you sample one bag, the odds of finding it in those 25 grams are only 9.5%.  Sampling 7 bags increases the odds of capture to over 50%.  It would take sampling 47 bags (4,700 contaminated seeds and 1175 gm sample) to increase the probability of capture to 99%.  So when you are testing a full truckload of seed (800 bags), contamination only needs to be in 47 (6%)of the bags, at very low levels, to get a 99% probability of capture.

But because the protocol is to draw at least 3 kg, sampling one bag, 120 times, would give you a 99.9+% chance of capture.  Sampling 7 bags with 17.15 pulls each (120 total) will also give you 99.9+%, or pulling 2 pulls from each of sixty bags would give you the 99.9+% chance.

Bags Per Lot

1

1

7

7

47

47

800

CS/kg

4

4

4

4

4

4

4

CS/lot

100

100

700

700

4700

4700

32000

Samples

1

120

7

120

120

120

800

Seed Sampled

25g

3kg

175g

3kg

3kg

3kg

20kg

Probability

9%

99.9+%

50%

99.9+

%

99.9+

%

99.9+

%

99.9+

%

Table 11. Even small lots can be sampled with a high probability of capture if the seed is contaminated.

The Effectiveness of ISS Seed Screening is Inversely Related the Effectiveness of Chlorine.

The more contaminated the seed is, the less effective seed sanitizing is.  Yet, the more contaminated the seed is, the easier it is to detect a pathogen using ISS’ Seed Screening Procedures.

sprout_seed_screening

Figure 12.  ISS Seed Screening is most effective when the seed is heavily contaminated, which is when sanitation is least effective.

If we compare two lots of seed; one with 1,000 contaminated seeds and one with 10,000 contaminated seeds.  In order to have a complete kill on the first lot, you have to kill the pathogens from 1,000 seeds.  In the second lot, you have to kill those same pathogens from 1,000 seeds, and kill the next 1,000 contaminated seeds, and the next and next until you have killed 10 times as many pathogens as the first group.  So chlorine has better odds of effectively sanitizing the 1,000 contaminated seeds than the 10,000 contaminated seeds.

On the other hand, seed sampling has a 63.2% chance of capturing a contaminated seed in the first lot, and a 99.9% chance in the second lot if the contamination is distributed evenly in the contaminated bags.

Seed screening along with decontamination complement each other to reduce the risk of salmonella or E.coli 0157:H7 in sprouted seed.