Annex 1: Additional Results from the BSE-Like/Non BSE-Like Test

Annex 1: Additional Results from the BSE-Like/Non BSE-Like Test

SE0241 Annexes to SID5

Annex 1: Additional Results from the BSE-like/non BSE-like Test

This annex contains further results from the BSE test, in addition to those given in Section 3 of the main text.

A1.1 IAH Data - analyses of single mouse lines

The distribution of sources by their incubation period and the first principal component of their relative lesion profile (73% of total variation) in RIII mice is shown in Figure A1.1. While one of the vCJD sources lies outside the 99% BSE-like prediction region, there is a degree of discrimination between the BSE-like (maximum LRS = 13.7) and non-BSE-like sources (minimum LRS = 22.1). Adding the second and third principal components from the relative lesion profile increased somewhat the discrimination between BSE-like (maximum LRS = 22.2, 4 d.f.) and non-BSE-like sources (minimum LRS = 63.5, 4 d.f.). An analysis based on relative lesion profiles alone, even when including the first 6 principal components (99% of total variation) failed to provide any discrimination between BSE-like (maximum LRS = 48.1, 6 d.f.) and non-BSE-like sources (minimum LRS = 10.9, 6 d.f.). An analysis of C57 mice alone produced a broadly similar degree of discrimination to that obtained with RIII mice (data not shown). Analyses of VM mice alone and of C57xVM mice alone provided poorer discrimination (data not shown).

Figure A1.1 Analysis of RIII Mice Only

A1.2 IAH Data – analysis of RIII and C57 mouse lines

Further analyses were performed to determine the extent to which combining data on RIII and C57 mice enhances our ability to discriminate between BSE-like and non-BSE-like sources. An analysis based on incubation periods alone provided some discrimination between BSE-like (maximum LRS= 18.1, 2 d.f.) and non-BSE-like sources (minimum LRS = 44.1), although one of the vCJD sources lay outside the 99% BSE prediction region (Figure A1.2). Including the first principal component of the relative lesion profile (Figure A1.3) also provided discrimination with all BSE-like sources lying within the 99% prediction region (maximum LRS = 8.1, 2 d.f., P = 0.02) while all non-BSE-like sources lie well outside the region (minimum LRS = 16.6, 2 d.f., P = 0.0002). Including additional components of the relative lesion profile did not lead to any substantial improvement in discrimination.

Figure A1.2 Analysis of RIII and C57 Mice – Incubation Period

Figure A1.3 Analysis of RIII and C57 Mice – IP Ratio and LP 1st Principal Component

A1.3 Analysis of VLA Data

Data were available on 419 transmissions into RIII, C57Bl or VM mice. In these analyses we focus on the transmissions into RIII and C57Bl mice and comparing the results with those from IAH. (Transmissions from natural BSE sources to VM mice were not performed.)

Type of source / Number of transmission to:
RIII mice / C57Bl mice / RIII and C57BL mice
Natural BSE in cattle / 52 / 36 / 27
Natural TSE in sheep / 256 / 166 / 147
Experimental BSE in sheep / 10 / 1 / 1
Experimental BSE in pigs / 0 / 6 / 0
Mixed BSE/scrapie / 4 / 4 / 4
Subpassaged source / 37 / 37 / 36
Total / 359 / 250 / 215

There were notable differences between the incubation periods observed in the VLA experiments and those observed at IAH for BSE sources. The median incubation period in RIII mice across the 52 natural BSE sources transmitted at VLA was 522 days. The typical incubation period in RIII mice at IAH is around 320 days. Similarly, the median incubation period for VLA natural BSE sources in C57Bl mice was 695 days compared with a figure of around 420 days at IAH.

Application of the results of IAH analyses to VLA data

Unsurprisingly, given the above, application of the IAH principal components analysis based on RIII mice alone to the VLA data resulted in serious misclassification of sources as BSE-like or not-BSE-like (Figure A1.4). The shift towards longer incubation periods for known BSE sources is evident, with most VLA BSE sources lying outside the IAH defined BSE-like region. At the same time a small number of natural sheep TSE sources lie inside the BSE-like region as do some sources undergoing subpassage.

Application of the IAH principal components analysis based on both RIII and C57 mice to the VLA data performs somewhat better (Figure A1.5), as the analysis is based on the ratio of incubation periods in C57 and RIII mice rather than the absolute length of the incubation periods. The majority, but by no means all, of the natural BSE sources lie within the IAH defined BSE-like area. However, several BSE sources lie well outside the area, while a number of natural sheep TSE sources lie inside or very close to the BSE-like area. The BSE sources lying outside the BSE-like area were from Experiment 1901, sources 641 (cervical spinal cord), 682 (thoracic DRG), 776 (frontal cortex), 864 (frontal cortex), 870 (trigeminal ganglia). Natural sheep sources lying within the BSE-like 99% prediction region were experiment number 1929 source 533 (spinal cord); experiment number 1945 sources 012, 014, 015, 016, 017, 023 (all brain), 0.29, 0.30 (rostral medulla); experiment number 1919 source 059 (brain); experiment number 1938, sources 068 and 073 (both brain). Increasing the number of principal components of the lesion profiles included in the analysis to as many as 5 does not resolve the problem with natural sheep sources still well inside the BSE-like region while BSE sources lie well outside.

Figure A1.4 VLA Data with IAH PCs – RIII Data

Figure A1.5 VLA Data with IAH PCs – RIII and C57 Data

Internal analysis of VLA data

An internal analysis of the VLA data was performed using a similar approach to that applied to the IAH data. For this analysis, data from Experiment SE1901 (transmissions involving natural, cattle BSE sources) were used to define BSE-like areas. Data from other experiments were then tested against these areas. The results of an analysis based on RIII mice only are shown in Figure A1.6. As expected the BSE sources used to define the BSE-like prediction region lie mostly within the region. One natural BSE source 1901/776, with data from only one mouse, lies outside the region. All nine of the experimental BSE in sheep sources with the required data lie within the BSE-like region but so too do many natural sheep TSE sources. Increasing the number of components of the lesion profile included in the analysis did not enable discrimination between BSE and natural sheep TSE sources.

Figure A1.6 VLA Data with VLA PCs – RIII Data

A similar analysis was performed based on C57BL mice alone. The results are shown in Figure A1.7. As expected the natural BSE sources lie inside the BSE-like prediction region as does the one experimental BSE in sheep source. However, many natural sheep TSE sources also lie within the BSE-like region. Increasing the number of lesion profile components used in the analysis up to 5 improves the situation but does not resolve this problem.

Figure A1.7 VLA Data with VLA PCs – C57 Data

A similar analysis was performed based on both RIII and C57Bl mice. The results are shown in Figure A1.8. Again, as expected, the natural BSE sources lie inside the BSE-like prediction region with the exception of 1901/776, as does the one experimental BSE in sheep source. However, many natural sheep TSE sources also lie within the BSE-like region. Increasing the number of lesion profile components used in the analysis up to 5 not resolve the problem.

Figure A1.8 VLA Data with VLA PCs – RIII and C57 Data

Finally, an analysis was performed using only lesion profile data from RIII and C57Bl mice (Figure A1.9). Excluding the incubation period data does not resolve the problem of distinguishing BSE sources from presumed non-BSE sources.

Figure A1.9 Figure A1.8 VLA Data with VLA PCs – Lesion Profiles Only

Annex 2 Testing for Gaps between Clusters

As discussed in the main body of the text (Section 4.2.4) we have chosen to define clusters such that two distinct clusters must have a clear gap between them. This definition has to be backed up with an objective criterion for the existence of a gap. Here we describe the criterion adopted, based on an analysis of the proximity matrix.

For any two items, A and B, in a data set on which we are doing cluster analysis, we have a distance d(A, B) between them. Another name for these distances is “proximities” (i.e. “how close are they together” rather than “how far are they apart”). Suppose then we have two putative clusters 1 and 2. Let us start by looking at cluster 1, and take all the pairwise proximities between its own members. This will define a distribution of intra-1 proximities, ranging from those between nearest neighbours out to those at opposite ends of the clusters. If there are N1 members of cluster 1, the number of intra-1 proximities will be N1(N1 – 1)/2 . The idea then is:

for there to be a gap between the clusters, the nearest 2-member to the 1-cluster must be significantly further away from the 1-cluster than the typical distance between near neighbours in the 1-cluster.

To turn this into a mathematical test, we have to make the terms precise:

nearest 2-member to the 1-cluster – to define this we look at all the proximities between 2-members and 1-members and find the smallest value; the 2-member with this value is defined to be the nearest to the 1-cluster;

distance between the nearest 2-member and the 1-cluster – this distance, which we call D21, we take to be the smallest value identified in the previous step, i.e. the distance between the nearest 2-member and the 1-member that is closest to it;

typical distance between near neighbours in the 1-cluster – to get a measure of this we take the first N1 proximities in the intra-1 distribution and express this as a percentile of the distribution (this value is 2/(N1 – 1) ).

We then express the distance D21 as a percentile of the intra-1 distribution. The larger this is compared to the percentile calculated in the third step above, the more significant is the gap between the clusters.

It is always possible that this nearest 2-member to the 1-cluster is something of an outlier from the 2-cluster that creates a “bridge” between the clusters. To look for that we eliminate this 2-member and repeat the process to find the second nearest 2-member, and then repeat this to find the third nearest. (This can be iterated further, but in the spreadsheet that automates this method we stop at the third nearest.) If they are all as close to their nearest 1-neighbour as 1-members are to their neighbours, we cannot say that there is a significant gap.

We then turn the test around and compare the distances of the three nearest 1-members to the 2-cluster with the intra-2 distribution. It is possible that a gap can seem large on one of these tests and not on the other. This can happen when one of the clusters (say cluster 1 is more diffuse than the other. The distance between the two nearest inter-cluster neighbours may be large compared with the distances between intra-2 neighbours, but may be comparable to the distances between intra-1 neighbours.

We now illustrate this method using the examples referred to in the main text.

A2.1 V Cluster Substructure

The first example of this analysis is for the supposed V1 and V2 subclusters in the data displayed on Figure 4 in the main text.

closest V2 members from V1 / closest V1members from V2
1st / 2nd / 3rd / 1st / 2nd / 3rd
item / 1919/008 / 1919/027 / 1919/058 / item / 1919/012 / 1919/039 / 1919/073
distance / 0.920 / 1.222 / 1.314 / distance / 0.920 / 1.222 / 1.269
%ile of V1 intras / 3.0% / 11.7% / 14.7% / %ile of V2 intras / 4.4% / 17.8% / 20.0%
compare with 1st n1 / 9.5% / compare with 1st n2 / 22.2%

Table A2.1 Proximity Analysis for V1 and V2

The left hand side of the table identifies the three V2 members closest to the V1 clusters, and compares the distances with the distance distribution within V1. The shortest distance across the divide is considerably smaller than typical distances within V1. The second and third are somewhat larger than the first n1 intra-V1 distances, but not substantially so. All of the three distances from the nearest V1 members to V2 are within the first n2 distances within V2. In other words, distances across the supposed gap are typical of the shorter distances within V2. From this we can conclude that there is no objective gap separating V1 and V2, and that therefore they should not be considered as distinct clusters.

A2.2BSE Substructure

Next we examine the apparent gap between the two BSE subclusters in Figure 5 in the main text, BSE1 (mostly brainstem or whole brain) and BSE2 (other brain, spinal cord, ganglia). The results of the analysis of proximities are as follows

closest BSE2 members from BSE1 / closest BSE1members from BSE2
1st / 2nd / 3rd / 1st / 2nd / 3rd
item / 1901/734 / 1901/868 / 1901/887 / item / 1901/184 / 1901/188 / 1901/187
distance / 0.474 / 0.643 / 0.731 / distance / 0.474 / 0.588 / 0.635
%ile of BSE1 intras / 8.3% / 23.3% / 35.0% / %ile of BSE2 intras / 3.6% / 7.5% / 12.3%
compare with 1st n1 / 13.3% / compare with 1st n2 / 9.1%

Table A2.2 Proximity Analysis for BSE1 and BSE2

Only one of the BSE2 members has a distance from BSE1 comparable with the typical intra-BSE1 distances. This is the inoculum 1901/734, already identified by visual inspection as a potential “bridging point”. Two of the BSE1 members are closer to BSE2 than the typical intra-BSE2 distance, but in both cases, these are distances to 1901/734. This suggests that when this one point is omitted, a well-defined gap should remain. This is borne out on Table A2.3.

closest BSE2 members from BSE1 / closest BSE1members from BSE2
1st / 2nd / 3rd / 1st / 2nd / 3rd
item / 1901/868 / 1901/887 / 1901/736 / item / 1901/188 / 1901/187 / 1901/190
distance / 0.643 / 0.731 / 0.846 / distance / 0.643 / 0.790 / 0.796
%ile of BSE1 intras / 22.7% / 37.9% / 51.5% / %ile of BSE2 intras / 13.9% / 35.9% / 36.8%
compare with 1st n1 / 18.2% / compare with 1st n2 / 9.5%

Table A2.3 Proximity Analysis for BSE1 and BSE2, Omitting 1901/734

Annex 3 The Absence of A-Cluster Members in SE1938

As noted in the main body of the text (end of Section 4.2.3),there are no SE1938 inocula with strain typing signals near the A-cluster. This is demonstrated on Figures A3.1 and A3.2 below. In these principal components analyses, the reference clusters were put together with the SE1938 results, for C57 and RIII mice respectively. The SE1938 points are distinguished according to whether the donor was an x/V (i.e. either A/V or V/V) sheep or an A/A sheep. To get as full a coverage as possible, even inocula with only one “full mouse” in the panel have been included.

Figure A3.1 Reference Clusters and SE1938 Scrapies – in C57 Mice

Figure A3.2 Reference Clusters and SE1938 Scrapies – in RIII Mice

The SE1938 points lie on or around the V cluster, and none are unequivocally members of the A cluster. There is more scatter in the SE1938 points than in those from the reference V clusters, but that is only to be expected, given that we are including here inocula with attack numbers as low as one mouse. An analysis of the RIII results shows that, indeed, the points furthest from the V cluster all correspond to attack numbers of four mice or less.

Since A-cluster signals were an important feature of the SE1919 results, and were also found in the IAH data, the question is: why are there no A-cluster members in the extensive body of natural scrapie data provided by experiment SE1938? Based on the earlier data we would have expected around half the A/A inocula (of which there are 9 in C57 and 11 in RIII for SE1938) to be close to the A-cluster, but in fact none are seen.

There were three differences in experimental design between SE1919 and SE1938, which might be relevant to this result.

  • The SE1919 sheep were collected between 1996 and 1999, whereas the SE1938 sheep were collected between 1998 and 2003.
  • In SE1919 the sheep were mostly collected individually from separate farms, whereas SE1938 had most sheep in same-flock groups.
  • In SE1919, the inocula were prepared from a section of brain stem rostral to the severed medulla, comprising the rostral medulla oblongata and pons (tissue code 7). In SE1938, eight of the 143 inocula were mixtures of tissue from five different brain areas (each from the same sheep), namely areas coded 1, 2, 4, 5 and 6. Only one was from a single brain area (area 5), and the rest were mixtures of three areas (2, 4 and 5).

One possibility is that the A-cluster scrapie had died out by the time SE1938 began. To test this, the SE1919 inocula were ordered as a time-series by means of the sheep PG numbers. The result was that the A-cluster cases are distributed across the time-range, except for the earliest times. There were four A-cluster cases in 1998 and 1999, when SE1938 was collecting sheep, including A/A sheep. A date effect therefore is implausible. A strain would not disappear in such a short period of time.

Because the SE1938 sample was made up of same-flock groups, the suggestion was made that the A/A sheep might have been in contact with x/V sheep in the same flock. This might then have predisposed them to have scrapie that gives rise to the V cluster, because they were more likely to have caught the disease from their x/V neighbours. When the data on the genotype and ownership of the sheep was examined, the same-flock groups were found to be uniform with respect to genotype. Five of the 14 groups had only A/A sheep, and the remaining eleven groups had only x/V sheep. No group had A/A and x/V mixed together. This does not of course preclude the possibility that the samples came from mixed flocks.

Attack rates in the SE1938 inoculations tended to be low. Moreover the A/A inocula gave disproportionately low attack rates. This is shown in Figure A3.3 below, which gives the distributions of the sum of the C57 and RIII attack numbers (defined as the number of “full mice” in the sub-panel).