Quantitative Morphology in Kidney Research
February 13-14, 2012 Conference Videos
Podocyte Number: Methods to Estimate Podocyte Number
Kevin Lemley, University of Southern California
Video Transcript
1
00:00:00,000 --> 00:00:09,433
JEFFREY KOPP: So with that I’d like to acknowledge, once again, the excellent help from my Co-Chair, Kevin Lemley, who again is the
2
00:00:09,433 --> 00:00:19,333
guiding force behind much of this and Kevin, as you all know, is Chief of Pediatric Nephrology at USC and has been doing podocyte biology in the
3
00:00:19,333 --> 00:00:29,033
lab and in animals and stereology for many years. I’ve learned that in his youth he was a competitive back-stroke swimmer and still swims five times a
4
00:00:29,033 --> 00:00:35,899
week, so that’s a nugget for you. Okay. Thank you, Kevin.
5
00:00:35,900 --> 00:00:43,933
KEVIN LEMLEY: So, I’m actually hoping you’re going to find this mildly redundant after John’s excellent introduction to stereology yesterday.
6
00:00:43,933 --> 00:00:51,966
Some of the concepts are going to come out the same and instead of glomerular number in the kidney what I’m going to be concentrating on is
7
00:00:51,966 --> 00:01:02,932
the other big theme of this meeting which is podocyte number; the commonalities here being number. So, I’m just going to review for you.
8
00:01:02,933 --> 00:01:14,666
We’re going to use a little bit of symbology here to make things easy, and when I’m using N here it’ll be the number of podocytes per glomerulus; V is
9
00:01:14,666 --> 00:01:23,866
the average glomerular tuft volume; N(V) is this numerical density, how many podocyte nuclei, which is how we count podocytes per tuft
10
00:01:23,866 --> 00:01:36,766
volume; and N(A) being in a single profile how many hits or profiles of podocyte nuclei you see per area. There are four methods I consider to
11
00:01:36,766 --> 00:01:49,832
estimate, that is to say not guess, podocyte number per glomerulus. The first one is the gold standard. You just cut through the entire
12
00:01:49,833 --> 00:01:59,433
glomerulus. As John was saying, when you have multiple sections you pretty much never have a question of what’s a podocyte nucleus. It gives
13
00:01:59,433 --> 00:02:09,866
you directly the number in that nucleus, so it’s not a population estimate, it gives you a number for one glomerulus, but as you can imagine it’s
14
00:02:09,866 --> 00:02:17,499
somewhat time-consuming and this is, although the gold standard, not the thing that is used very often. The Weibel-Gomez method is a model-
15
00:02:17,500 --> 00:02:24,700
based method. As you remember, John said we have design-based methods and model-based methods, and model-based methods are kind of
16
00:02:24,700 --> 00:02:35,466
the original ones and this has seemingly been supplanted but I’m going to mostly spend my time this morning talking about it and what advantages
17
00:02:35,466 --> 00:02:44,199
it has. The “thick-thin” method has been described multiple times since the 20s with Wicksell and Abercrombie, more recently in called
18
00:02:44,200 --> 00:03:00,233
thick-thin in Roger Wiggins’ group earlier in this century, so we’re now spanning centuries with it and it gives you…I’m sorry…the thing I’d say...the
19
00:03:00,233 --> 00:03:06,233
Weibel method when you use it essentially gives you a density. It does not give you N, so that’s one difference between these; it gives you a
20
00:03:06,233 --> 00:03:15,933
density so you need to multiply by another population estimate of the tuft volume to get the actual N. Thick-thin also gives you an N. We’ve
21
00:03:15,933 --> 00:03:24,733
had a bit of debate over this. In my opinion the thick-thin is a design-based method and involves only measurements, no model assumptions. And
22
00:03:24,733 --> 00:03:34,633
the disector-fractionator, if you break out the disector which was first described, it gives you N(V), the density, and if you combine it with a
23
00:03:34,633 --> 00:03:43,533
fractionator it gives you directly N. So we have all of these combinations of different methods, one of which is gold standard, one of which is model-
24
00:03:43,533 --> 00:03:54,666
based, and two of which you could call design-based. One method that should not be used but is unfortunately is the most commonly used method
25
00:03:54,666 --> 00:04:04,432
is to look at the number of podocyte nuclei per tuft cross-section and call that the podocyte number. Often, as if it were the podocyte number
26
00:04:04,433 --> 00:04:12,399
per glomerulus, although when in the mouse you see the number as 8 or 10 or a human you see it’s 20 or so, you pretty much know it’s not really
27
00:04:12,400 --> 00:04:24,200
the podocyte number. And so, this is the number of WT1-positive cells from 50 random glomeruli per mouse and you can see these authors are
28
00:04:24,200 --> 00:04:34,100
showing that in the wild-type mouse you had a certain number and in an outport model of the mouse you lose podocytes and then if you
29
00:04:34,100 --> 00:04:43,866
reconstitute the out-port with amniotic fluid stem cells you actually repair this. It looks really great but we’re going to show why it’s maybe not so
30
00:04:43,866 --> 00:04:54,166
great, and you may think that I’m making fun of Sedrakyan but let me tell you one of the names of the authors on this paper: Lemley. So, it is not
31
00:04:54,166 --> 00:05:03,266
easy even in your own laboratory to get things done the way you would. Why is it that we don’t want to use this method? Again, you can see it
32
00:05:03,266 --> 00:05:13,266
shows a phenomenon; it is not that it tells you nothing. But if we for the time being take the Weibel estimate of N(V) as a gold standard just
33
00:05:13,266 --> 00:05:23,499
for relative comparison, I looked at a number of different things if you wanted to predict N(V). So, here’s Type II diabetic Pima Indians and this is just
34
00:05:23,500 --> 00:05:34,733
N(A) from the Weibel estimate. So if all you do is the simple part of the Weibel estimate you can see you basically resurrect almost all the
35
00:05:34,733 --> 00:05:43,533
information from Weibel, whereas if this is from the same data, you look at the number of podocyte nuclear hits per profile and use that as
36
00:05:43,533 --> 00:05:54,899
your estimator, you do get a relationship but you’ve lost an awful lot on the R value. Same thing for normal kidney donors; something we’ve
37
00:05:54,900 --> 00:06:03,000
been doing for quite a while. If you look at the N(A) component you see it basically replicates almost all of the information content of the N(V)
38
00:06:03,000 --> 00:06:13,733
from Weibel, R value at .96, whereas there’s still some relationship but it’s looking much more like a cloud if all you do is look at the number of nuclear
39
00:06:13,733 --> 00:06:23,666
hits per profile. IgA nephropathy…I’m not trying to bore you here but we’re seeing in various different situations you really get the same
40
00:06:23,666 --> 00:06:32,666
phenomenon, that you get a very high R value and a pretty good predictor, and here again, a little bit more like a cloud. And then even in the
41
00:06:32,666 --> 00:06:44,399
lowly rat the same phenomenon, a very high R value, a very good replication of the N(V) information content in N(A) but N(P) degrades it a
42
00:06:44,400 --> 00:06:57,300
fair amount. So again, I told you what I want to do is go over this model-based method considered not modern, shouldn’t be used, but show just
43
00:06:57,300 --> 00:07:04,500
some aspects of what we understand about it and what we don’t understand about it; in essence, any advantages and any limitations.
44
00:07:04,500 --> 00:07:14,266
Well, any single-section method…people who work with human material have one thing they can say about it, you can do it in your biopsy
45
00:07:14,266 --> 00:07:24,666
material, and that’s something that unfortunately does not by and large apply to design-based methods; animals, yes, autopsy material, yes. But
46
00:07:24,666 --> 00:07:35,132
actually biopsy material from living people, we’re usually pretty much limited to single sections and I’m on the Pathology Committee of the NEPTUNE
47
00:07:35,133 --> 00:07:46,433
Study where we’re doing digital microscopy on this. We’re trying to get various levels. What you get from centers—23 different centers—for their
48
00:07:46,433 --> 00:07:52,599
representation of levels there is no way you can put things together. You can’t reconstruct something that wasn’t done design-based from
49
00:07:52,600 --> 00:08:02,933
the start. So it’s a major advantage, again, in people who are concerned with human beings. It does not require any knowledge of section
50
00:08:02,933 --> 00:08:11,699
separation, which is required both in the thick-thin and the disector; you have to have the level to which you have imprecision in your knowledge of
51
00:08:11,700 --> 00:08:20,166
how far the sections are apart, is going to find itself right into your final answer. It does not require, as in the disector-fractionator which gets
52
00:08:20,166 --> 00:08:28,166
away from the N(V) thing, you do have to section through the entire object of interest. So if you’re trying to say, “I want to know how many
53
00:08:28,166 --> 00:08:39,566
podocytes there are in a glomerulus,” you must have the whole glomerulus represented in your fractionator sampling. So all of these have rather
54
00:08:39,566 --> 00:08:47,932
more work involved and it’s very controlled work that you really have to know what you’re doing; you can’t lose sections, you can’t do anything. So
55
00:08:47,933 --> 00:08:57,433
those are major disadvantages when doing it in human material where you don’t get another bite at the apple. Again, as a disadvantage, you only
56
00:08:57,433 --> 00:09:07,233
get this density and so you have to get at the same time an estimate of the average glomerular tuft volume and you only, therefore, get a
57
00:09:07,233 --> 00:09:14,599
population mean. The population could be the individual but it isn’t the individual glomerulus. You’re not going to be able to get, “Oh, in this
58
00:09:14,600 --> 00:09:25,966
glomerulus the podocyte number is X.” It’s going to be: in this individual the average podocyte number is X. All single-section methods have a
59
00:09:25,966 --> 00:09:33,466
common problem and this may look, again, now a bit redundant after hearing John’s excellent talk yesterday and the fundamental problem is not
60
00:09:33,466 --> 00:09:43,166
that your density you see in a single cut-through doesn’t have a relationship to N(V), which is what we’re after in this case, it’s that it’s
61
00:09:43,166 --> 00:09:51,632
influenced by a number of other things, and in particular by how big the particle is and we refer to that often as the caliper diameter, which is a
62
00:09:51,633 --> 00:10:00,566
statistical meld as if you looked at it from every direction, what would be the largest diameter you would have. So, it’s something that
63
00:10:00,566 --> 00:10:11,966
accommodates non-uniform sized particles but it really influences it. So way back in the 40s Abercrombie showed that…and the section
64
00:10:11,966 --> 00:10:20,232
thickness also had an effect, you remember, and it has to do also with the probability of hitting a particle. So the simplest version, if you’re
65
00:10:20,233 --> 00:10:29,333
assuming that the section thickness is much less than your particle diameter would be—not that N(A)…you remember John had this non-
66
00:10:29,333 --> 00:10:39,099
equations going across, N does not equal N(A) does not equal N(V)—but if you make this assumption about “t” and you know D, which is
67
00:10:39,100 --> 00:10:48,966
of course where the hitch is, then N(V) equals N(A) divided by D. So you can reconstruct it, you just need another piece of information; it isn’t
68
00:10:48,966 --> 00:11:02,966
automatic in the N(A) to get N(V), you need the D. So this is the Weibel formula, N(V) being your density of particles in a volume, N(A) is the areal
69
00:11:02,966 --> 00:11:16,732
density of particle hits, and A(A) is the absolute areal percentage of your reference area that is due to your particle cross-sections. So those are
70
00:11:16,733 --> 00:11:25,066
two measured things and this is your parameter; this is a shape factor. In essence, what it is from the original Weibel-Gomez derivation is a
71
00:11:25,066 --> 00:11:34,699
dimensionless ratio of the particle volume average to the mean particle cross-sectional area. So if you can get that in some way, you can
72
00:11:34,700 --> 00:11:43,766
get beta. There is often, on the top here, a kappa which takes into account the fact that not all your particles are the same size, so we’re talking here
73
00:11:43,766 --> 00:11:53,532
about the effect of particle shape and here about that it’s missing the effect of particle size and what it turns out and usual practice and has been
74
00:11:53,533 --> 00:12:04,066
shown in John and Wendy’s group with respect to glomerular estimation, that over normal populations with a fair amount of variability, in this
75
00:12:04,066 --> 00:12:13,032
case from their work in glomerular size, you’re talking about 1 to 1.03, that it really is…if you just want to say “1” you’re not going to be off by very
76
00:12:13,033 --> 00:12:24,599
much. So, this gets suppressed or ignored a number of times. So what I want to do is look at this—I hate to use the word because it’s kind of
77
00:12:24,600 --> 00:12:34,700
biased—this biased method, I like model-based, and see how sensitive it is to these assumptions you need to use in order to get a value out of it
78
00:12:34,700 --> 00:12:43,333
because there’s no question, you have some assumptions you’re making, and in particular since we’ve said that kappa probably can be
79
00:12:43,333 --> 00:12:51,133
ignored because it’s only a couple of a percent until you get a lot of variation, how sensitive is it to this beta factor, because everything else in
80
00:12:51,133 --> 00:13:00,833
there is a measurement? So, I want to use something called sensitivity analysis which is to look at the behavior of a model on some of its
81
00:13:00,833 --> 00:13:09,133
component parameters that go into making the model work. There’s a form of it called local sensitivity analysis that looks at what happens
82
00:13:09,133 --> 00:13:17,066
when you look around a nominal value for one of the parameters and say if we vary that, what happens to the outcome? How much does it
83
00:13:17,066 --> 00:13:26,532
affect it? If in fact you’re looking at the performance of something like a formula, like the Weibel formula, you can do this analytically. You
84
00:13:26,533 --> 00:13:35,666
actually can solve it and say how sensitive is it to this parameter; it’s not an empirical thing. And so, if you get a formula which connects up the
85
00:13:35,666 --> 00:13:43,299
variables including the parameters, you basically can just look at the partial derivative with respect to any one variable and then say, okay, how
86
00:13:43,300 --> 00:13:52,200
much is it changing with the change in that parameter? In this case N(V) is the outcome variable and we’re going to look at beta as the
87
00:13:52,200 --> 00:14:04,066
input. To do that let’s look at, not again some sort of odd-shaped particle, but something where we can actually get the beta, that’s easy to calculate,
88
00:14:04,066 --> 00:14:13,566
and say that, well, we’re looking at something that deviates from spherical. So this again might be a better glom example, but if you just pull out
89
00:14:13,566 --> 00:14:24,299
one of the axes of a sphere—so, if it’s originally A and A, that would be a sphere—and you make B longer, then that becomes an ellipsoid. And so we
90
00:14:24,300 --> 00:14:33,700
can do this analytically for ellipsoids just as an example, and in fact in what will be shown here, let’s just look at the ratio between the long axis to
91
00:14:33,700 --> 00:14:46,566
the short axis and call it lambda. So if you look at lambda going from spherical to fairly pulled out, the shape factor—and I apologize, I only realized
92
00:14:46,566 --> 00:14:58,732
after I got here and could no longer change them on my NCSS program that this is off here—but basically these are from .3 to .5 and so you get
93
00:14:58,733 --> 00:15:09,866
not a big change in the shape factor over a pretty large amount of elongation. And just to show you what values you get…so this is, unfortunately,
94
00:15:09,866 --> 00:15:19,766
the derivatives go to infinity if you use 1, so if you look at something that’s just almost a sphere that you couldn’t tell the difference, you get this
95
00:15:19,766 --> 00:15:30,232
canonical beta value which you heard from John already of 1.382 and what you get is something where basically there is almost no change as you
96
00:15:30,233 --> 00:15:40,833
start elongating. So down at a spherical size, the shape factor doesn’t change at all as you elongate. It would be 0 if this were 1, but you see
97
00:15:40,833 --> 00:15:49,399
it’s a very small number. These are actually drawn to these axes, so this is just having a quarter more length than you have the sideways
98
00:15:49,400 --> 00:16:01,966
dimension. You get about a 1.4% increase in your beta for a 25% elongation and that’s represented by the fact that, really, you’ve got a pretty small
99
00:16:01,966 --> 00:16:11,199
number here. Per percent change in this ratio you get a much lower percent change in beta; it’s not very sensitive. By the time you get up to
100
00:16:11,200 --> 00:16:20,733
something here, which really doesn’t look much like a sphere anymore—it’s really off—you’re still only talking in your shape factor of only a 4%
101
00:16:20,733 --> 00:16:31,299
effect, and again, only about 20% that comes from this. So in terms of the beta factor, it really looks like the Weibel formula is not terribly
102
00:16:31,300 --> 00:16:40,033
sensitive to what we might call uncertainty in it, the fact that in any given population we really don’t know what it is within some fairly broad
103
00:16:40,033 --> 00:16:49,899
parameters. So, how else can tell how much to trust the Weibel-Gomez method? Again, I’ve shown that in terms of its parameters which
104
00:16:49,900 --> 00:16:57,766
have been held against it as a model-based method, it’s not real sensitive the parameters. It doesn’t care too much for that parameter which
105
00:16:57,766 --> 00:17:07,999
is pretty much the main parameter. Is there any other way we can tell how to trust it? Well my friend, Mr. Basgen, has given us one of the rare
106
00:17:08,000 --> 00:17:18,800
examples of getting a gold standard and comparing different things to it. So, this is one of the more important papers. I don’t know, John, did
107
00:17:18,800 --> 00:17:28,600
you refer to it? Okay. It is in your thing. I recommend reading it because it’s actually looking at someplace where you know in an animal it’s
108
00:17:28,600 --> 00:17:40,166
only nine gloms in a mouse but you have a true standard and you can try different methods on it. And so, just to be sure when you read it, they
109
00:17:40,166 --> 00:17:48,999
weren’t counting podocyte number, they were counting total nuclear numbers—a total cell number—and looking at three methods: the
110
00:17:49,000 --> 00:17:57,733
Weibel-Gomez, disector-fractionator, and then the complete enumeration is the gold standard for the true number in there. For the Weibel-Gomez
111
00:17:57,733 --> 00:18:06,533
method, unlike the usual way it’s done where you have one cut-through, they did use multiple sections but the fact is, that also represents the
112
00:18:06,533 --> 00:18:15,499
fact the disector-fractionator is on multiple, so it kind of makes them on a little more even keel and they didn’t use a mouse value for beta, they used
113
00:18:15,500 --> 00:18:24,266
something taken from human but I don’t think anybody has a mouse value for beta. Here are the results of the enumeration; the actual for the
114
00:18:24,266 --> 00:18:37,432
nine glomeruli from the mouse. They had 213 cells per glomerulus on average, the disector-fractionator gave 211, and Weibel was off by a
115
00:18:37,433 --> 00:18:47,666
bit and was high by maybe 10% or so. So I like to plot things and not look for averages, so let’s look at what happens when we plot it. So here’s the
116
00:18:47,666 --> 00:18:57,499
gold standard; here’s the actual number. These are the individual nine mouses or nine glomeruli, and the two estimates; the Weibel is the red
117
00:18:57,500 --> 00:19:08,566
triangle and disector-fractionator is blue. So, one thing that’s pretty clear at the beginning is that the dispersion of the disector-fractionator is actually
118
00:19:08,566 --> 00:19:16,832
a fair amount more than Weibel and that’s represented in the fact that the R-squared, the coefficient of determination, is substantially
119
00:19:16,833 --> 00:19:24,033
higher in the Weibel than the disector-fractionator, basically just saying the numbers are tighter and you remember—we’re going back once more to
120
00:19:24,033 --> 00:19:33,833
John’s targets—you can have something that is right-on but broad or something that may be on or may be off but is tighter. In this case we’d say the
121
00:19:33,833 --> 00:19:44,199
Weibel’s a bit tighter than the disector-fractionator and then if we look at one other thing we see the slope of the relationship between the true and the
122
00:19:44,200 --> 00:19:54,433
Weibel and it’s just about 1, which is often what we would like for an estimator to be—intercept is 25, neither of them go to 0 at that—whereas the
123
00:19:54,433 --> 00:20:03,333
slope of the disector-fractionator is off a bit. Even though the means seem to hit right, kind of as a theoretical predictor, it’s both tighter and it seems
124
00:20:03,333 --> 00:20:16,433
to have a much tighter relationship to the true enumeration value. So in this example, Weibel-Gomez seems to both have higher precision and
125
00:20:16,433 --> 00:20:28,133
a slope near unity as a predictor of the enumeration values, so it doesn’t seem to be falling quite on its face. I was kind of surprised
126
00:20:28,133 --> 00:20:35,766
when I plotted it out that way; it looks a little better than what the averages would tell you. The average would say 211-213 this is great, but as a
127
00:20:35,766 --> 00:20:46,799
predictor it seems to have maybe some advantages, even. If you remember those plots again, looking in the Weibel N(A) showing this
128
00:20:46,800 --> 00:20:54,366
highly linear relationship to N(V), again they’re from the same measurement, so all it’s doing is saying if you look at part of the information that
129
00:20:54,366 --> 00:21:03,532
goes into Weibel, you really can reconstruct all of Weibel and why should N(A) be such a good predictor of N(V)? You’ve got the A(A) in there
130
00:21:03,533 --> 00:21:13,533
that you’re not using. So here’s the formula, but let’s say, in fact, that in human material you have lots of hits on podocyte nuclei and let’s say
131
00:21:13,533 --> 00:21:22,066
podocyte nuclei are actually relatively homogeneous and you do cross-sections through them. You may actually get a
132
00:21:22,066 --> 00:21:34,766
cross-sectional area that doesn’t have much variance, that’s actually kind of constant, and if you look at the cross-sectional area times N(A) it
133
00:21:34,766 --> 00:21:46,632
actually is A(A). And so if you just put those in and say that I can convert this to N(A), then you see that one of the N(A)s cancels here, we can
134
00:21:46,633 --> 00:21:55,399
take the beta out, the square root of a square is just N(A) and what you see is, in that type of assumption, that there’s a homogeneity among
135
00:21:55,400 --> 00:22:07,966
the glomeruli, so their cross-sectional areas tend to have a low dispersion that you would expect that N(A) is very close to linear-related to N(V).
136
00:22:07,966 --> 00:22:17,199
So, are there any other comparisons we have of Weibel-Gomez estimates of podocyte counting to a gold standard like complete enumeration? I don’t
137
00:22:17,200 --> 00:22:25,300
know. Maybe someone out there in the stereology land knows. I just don’t know of any; that’s why the Basgen paper is so important. It’s a lot of
138
00:22:25,300 --> 00:22:31,266
work but you actually get something out of it because you have the true gold standard you can compare with, but we really are kind of
139
00:22:31,266 --> 00:22:43,366
stuck. Now Kath, who’s in the back there and her colleague, Rudy Bilous, have done a number of very good studies and they looked at how
140
00:22:43,366 --> 00:22:53,532
comparable you can get podocyte number in different studies, and let’s just do a little personal humiliation here. You notice there’s an outlier but
141
00:22:53,533 --> 00:23:02,999
remember from what was said yesterday at that point we were using EM resin-embedded material to get our density and paraffin-embedded material
142
00:23:03,000 --> 00:23:12,266
to get our glomerular tuft volume, and we were accounting for perfusion and lack thereof and paraffin shrinkage, and even so, I think we got it
143
00:23:12,266 --> 00:23:19,899
wrong, and if you looked at it and it was in one of the tables yesterday, the average tuft volume—our tuft volume—was a lot smaller than other
144
00:23:19,900 --> 00:23:31,466
studies. So otherwise, you’ve got this 500-800 which is a decent agreement. We have now switched to doing everything in Epon which does
145
00:23:31,466 --> 00:23:40,532
make the tuft volume estimate a little harder because you can use a nice, big paraffin core and get many more glomerular cross-sections
146
00:23:40,533 --> 00:23:49,399
than the little bitty Epon cores, but you don’t have a shrinkage problem anymore and strange to say, in intraoperative biopsy of transplant donors,
147
00:23:49,400 --> 00:23:58,933
which we have 120 but we’ve done morphometry on 55, we get a value right in the middle if you just do the tuft volume right; it gets a
148
00:23:58,933 --> 00:24:12,199
lot better. So, here is another interesting study by Rudy and Kath and if you look at using the disector-fractionator and you can pick anything,
149
00:24:12,200 --> 00:24:26,666
let’s pick controls just for example, 580 was their average in the 10 controls and if you used the Weibel on electron micrographic images for the
150
00:24:26,666 --> 00:24:38,299
density you get something that is very close to 580. If you use Weibel in light it doesn’t do so well, but remember again that Abercrombie
151
00:24:38,300 --> 00:24:48,666
formula. The thickness of the section comes in to how well your estimator works and so when you get down to electron micrographic thickness,
152
00:24:48,666 --> 00:24:57,832
apparently you really eliminate that effect and it gets better, whereas if you get thick sections you basically make Weibel not work. And so, what
153
00:24:57,833 --> 00:25:09,199
they showed was Weibel actually in all of these cases—this is Type I diabetes patients in two different times—475, 467 at EM level Weibel
154
00:25:09,200 --> 00:25:17,200
actually does pretty well with the parameters that are used. You’ve got the problem that EM level is pretty difficult. Maybe it’s not more difficult than
155
00:25:17,200 --> 00:25:27,033
doing a disector fractionator which has a lot of comparisons and measurements, but it’s not like doing light; EM is a lot more difficult. So what can
156
00:25:27,033 --> 00:25:35,833
we say about bias in the Weibel-Gomez method? Well, bias comes into any estimator if the assumptions of the model don’t, in fact, obtain in
157
00:25:35,833 --> 00:25:48,733
reality. That’s probably not a big surprise. Weibel has several assumptions which could be a cause of bias. Unfortunately, through all of this because
158
00:25:48,733 --> 00:25:58,433
we think of Weibel as a formula—you should go back and read Weibel and Gomez’s original paper from the 60s, it’s actually a brilliant paper, but they
159
00:25:58,433 --> 00:26:05,966
list their assumptions—there’s one assumption that we never mention anymore and that’s that the model requires a constant number of
160
00:26:05,966 --> 00:26:13,899
transections per unit area of section. That doesn’t show up in the formula anywhere; it is an assumption explicitly stated by Weibel and
161
00:26:13,900 --> 00:26:23,433
Gomez. However, this is manifestly not the case for podocytes. We may apply it there but we’re already starting, even though it isn’t showing up
162
00:26:23,433 --> 00:26:31,866
in a formula, with something that is not the case. In fact, as we all know from looking at podocytes, they tend to be clustered, they’ve got sticking in
163
00:26:31,866 --> 00:26:41,699
the middle of a glomerular tuft, they tend to be differentially distributed toward the outside of the tuft. And so if you look at N(A) in the smaller
164
00:26:41,700 --> 00:26:51,466
N(A)s—the peripheral sections—you’re going to have a much higher N(A). So I want to raise a question: is it possible that the main source of
165
00:26:51,466 --> 00:27:00,432
bias in the Weibel-Gomez method really doesn’t have anything to do with nuclear size distribution, which in the glomerular case we’ve shown has a
166
00:27:00,433 --> 00:27:12,099
very small percent effect for 15% variation in diameter; or shape which again, remember, these rather ellipsoid-looking things really, just a few
167
00:27:12,100 --> 00:27:21,166
percent, maybe the fact that Weibel isn’t giving the right number is more related to the fact that we don’t have a homogeneous distribution of the
168
00:27:21,166 --> 00:27:35,866
nuclei throughout the whole glomerular tuft. Can we look at this? Well, my student in Heidelberg many years ago, made a little mistake when we
169
00:27:35,866 --> 00:27:47,799
were doing some stereology— Sibylle Tenschert—and I made her go back and do things. She had basically looked at glomerular profiles and taken
170
00:27:47,800 --> 00:28:00,166
the big, juicy round ones and done morphometry on them rather than any hit on a tuft counts. If you’ve got two capillary lumina and no podocytes
171
00:28:00,166 --> 00:28:08,232
and it’s random and you’ve picked it from a numbering system as a random hit, it counts, it goes into the database and if you don’t do that
172
00:28:08,233 --> 00:28:15,166
you’re doing biased sampling, and biased sampling is a real problem because once you’ve done biased sampling you can never reconstruct
173
00:28:15,166 --> 00:28:24,532
it; you have to be suspicious of it. I got suspicious and I asked young Sibylle to go back and do it over the right way. What we got from this was a
174
00:28:24,533 --> 00:28:36,433
wonderful database. We got a huge spread—this is in rats that had had a subtotal nephrectomy—and we got tuft areas from 1,000 square microns
175
00:28:36,433 --> 00:28:48,433
up to 44,000. So, we actually got a big spread and here’s what we see. As a function of tuft area, N(A) goes up as tuft area goes down. The
176
00:28:48,433 --> 00:28:55,566
more peripheral you get with your hit, the greater number of podocytes you’re going to hit; again, a reflection of the fact that they’re not uniformly
177
00:28:55,566 --> 00:29:05,599
distributed through the tuft. Looking at some data from IgA nephropathy and now humans that we’re occasionally interested in, we see exactly
178
00:29:05,600 --> 00:29:15,633
the same phenomenon. In fact here it looks like it’s a very linear relationship. That is, you go down in tuft area hits, you get more peripheral, N(A) goes
179
00:29:15,633 --> 00:29:27,433
up because the podocytes sit not uniformly but they are biased to sit on the outside of the tuft. Interestingly, if you look at endothelial-mesangial
180
00:29:27,433 --> 00:29:34,999
cells, and just like John, we really don’t think we can distinguish them, so we tend to throw them together in our studies. So in the normal kidney
181
00:29:35,000 --> 00:29:45,533
donors we got the same effect, that N(A) versus tuft area, highly significant, negative relationship, no relationship in the endothelium is there
182
00:29:45,533 --> 00:29:54,733
because they are more uniformly spread. In IgA where we showed that there’s this big negative relationship, no significant relationship to that cell.
183
00:29:54,733 --> 00:30:05,766
So it’s really the podocytes that kind of like to sit like kings on the outside of the glomerular tuft. So possibly, if we could actually take into account of
184
00:30:05,766 --> 00:30:14,932
the model the actual distribution of podocyte nuclei, we might be able to do something about this bias that I’m suggesting may be the majority
185
00:30:14,933 --> 00:30:23,966
of the bias you get in the Weibel-Gomez method and ideally, as you saw in the database from the IgA, we might actually be able to extract that from
186
00:30:23,966 --> 00:30:31,999
the very population we’re studying so that we can actually measure the correction in the population rather than say, “I’m going to use a
187
00:30:32,000 --> 00:30:45,000
beta of 1.55 because that’s what Steffes found some years ago.” And so, can we estimate a radial distribution of podocyte nuclei from the
188
00:30:45,000 --> 00:30:54,400
N(A) versus cross-section area, which again in the IgA patients really looked quite linear, and you notice it also had a very high R value in the
189
00:30:54,400 --> 00:31:02,966
transplant patients. Is it possible we can measure that? Or can we develop some sort of function that relates these parameters so we can get yet
190
00:31:02,966 --> 00:31:13,599
another correction factor for the Weibel formula, but again, one that’s based on things we actually measure in the population we’re trying to study?
191
00:31:13,600 --> 00:31:21,666
An alternative way of seeing this is in fact we’re sampling in a bad way, that we’re giving the same weight to a big sample of the glomerulus as we
192
00:31:21,666 --> 00:31:29,899
give to a little sample of the glomerulus, and that if we just weighted our N(A) estimates and didn’t just take them as a given for each section by the
193
00:31:29,900 --> 00:31:41,433
cross-sectional area we can maybe correct this. So, one thing I like to do is Monte Carlo methods and so we can make a synthetic system because
194
00:31:41,433 --> 00:31:50,066
not everybody’s going to have the diligence John Basgen had of actually going through using this very time-consuming method and getting a gold
195
00:31:50,066 --> 00:32:01,599
standard in biological material, but you can make up material. So, I had a student friend at Cal Tech do some modeling—I did the modeling, he did the
196
00:32:01,600 --> 00:32:11,300
programming—where we looked at 10 spherical reference bases with a little bit of built-in variation—how big the spaces were—embedded with
197
00:32:11,300 --> 00:32:25,600
100+/- 5 spherical particles. So, you notice here at the beginning we’ve gotten rid of size variation in the nuclei, if we’ll call the spherical particles
198
00:32:25,600 --> 00:32:33,400
nuclei, and we’ve gotten rid of shape variation. They’re all spheres and they may vary in number but they’re all the same size, so we’re only
199
00:32:33,400 --> 00:32:44,833
looking at their distribution. We did two types of embeddings. A normalized linear…this doesn’t look very linear, I realize, but the fact is that if you
200
00:32:44,833 --> 00:32:52,599
look at the volume element of the shell of the sphere as you go out, it actually goes up by the square with the linear dimension out just because
201
00:32:52,600 --> 00:32:59,900
its area is expanding. So this just actually gives you a linear and I’ll show you evidence of this, and this gives a normalized quadratic, so all of
202
00:32:59,900 --> 00:33:10,633
these numbers here are just to normalize it so you can put 100 particles into a sphere easily. So, here is 25,000 simulations going out from the
203
00:33:10,633 --> 00:33:22,866
radial dimension and this is your particle density per volumetric element, so this to the degree of our random sampling is a linear increase with the
204
00:33:22,866 --> 00:33:31,199
density—it goes up linearly as you go out—and then this is our represented of a quadratic increase, it goes up kind of more than linear; just
205
00:33:31,200 --> 00:33:41,433
as two examples. And so looking at the quadratic, here is the true N(V), because again in this case, we didn’t do it by enumeration but we know how
206
00:33:41,433 --> 00:33:48,833
many particles were in there because we put them, and this is what the Weibel method estimates with a couple of sections through. You
207
00:33:48,833 --> 00:33:57,666
can see we have a positive bias, and if you remember in John’s work the Weibel actually had positive bias in these nine mouse glomeruli. We’re
208
00:33:57,666 --> 00:34:05,566
off the line of identity and, interestingly, I’m not claiming that this is anything other than chance, remember the mouse had a coefficient of
209
00:34:05,566 --> 00:34:17,532
determination of .69 and this .63, that’s maybe dumb luck. So, if we use this model system and then we randomly sample 3 intersection levels in
210
00:34:17,533 --> 00:34:28,233
each of the 10 reference volumes, this was what turned out to be the density, and the fact is anybody who has done on human material
211
00:34:28,233 --> 00:34:36,666
stereology estimating with Weibel podocyte density, this is a good value for that. So, these values were chosen to give the type of particle
212
00:34:36,666 --> 00:34:46,666
per volume you get in human material. The unweighted usual way we do it showed a 9% positive bias, strangely again probably by
213
00:34:46,666 --> 00:34:54,699
chance, pretty much what John found in the mouse, but if you weight by the cross-sectional area you get something that’s a little bit closer:
214
00:34:54,700 --> 00:35:07,266
2.51 versus 2.49. Then if you do 4 replicates of this in each of the 10 systems using this quadratic density versus a straight just one-time-
215
00:35:07,266 --> 00:35:16,132
through Weibel thing, you get an 18% relationship—remember this is statistical—between your measured Weibel-Gomez N(V) and the true (N(V)
216
00:35:16,133 --> 00:35:26,033
but if you use the adjusted Weibel weighting it for the cross-sectional area, the fact is you get something that’s not different from 1. So, this
217
00:35:26,033 --> 00:35:35,799
approach seems in this synthetic system actually to be effective at eliminating what, by chance, may actually be the major form of bias in the
218
00:35:35,800 --> 00:35:49,200
Weibel method. So, what’s the main point of all of these mathematical hand-wavings? Single-section methods like Weibel, unfortunately biased,
219
00:35:49,200 --> 00:35:58,533
model-based as they are, are very likely to be the only practical methods we can use in routine pathologic material. We cannot do design on
220
00:35:58,533 --> 00:36:08,266
pathologic material even in a study, again this is becoming abundantly clear in the NEPTUNE study; you really have very little control off of what you
221
00:36:08,266 --> 00:36:17,099
get. It really pretty much, even in a study, looks like what you get from a standard clinical pathology lab. So, we may be stuck with Weibel
222
00:36:17,100 --> 00:36:28,033
or not studying humans. Bias can go hand-in-hand with a high precision method; again, as in the little bulls-eye thing that John showed. Some
223
00:36:28,033 --> 00:36:35,233
sources of bias, in particular with respect to Weibel, a precise knowledge of the shape factor may actually be a theoretical source of bias but
224
00:36:35,233 --> 00:36:43,799
functionally not very important; not really have much of an influence. Other sources such as I am proposing—the non-uniform distribution of the
225
00:36:43,800 --> 00:36:53,633
particles in the tuft—actually have not been well explored, even though they’re explicitly stated in Weibel, clearly violated in the tuft, but nobody’s
226
00:36:53,633 --> 00:37:05,933
really spent much time looking at them and it’s possible that they are subject to mitigation strategies. What remains to be done? Gold
227
00:37:05,933 --> 00:37:14,133
standard measurements such as complete enumeration should be used more often. People should make the effort. I think Kath said this,
228
00:37:14,133 --> 00:37:23,033
right? Sometimes if you put a lot of effort in but you get something out, maybe it’s worth doing the effort and I think that the Basgen study was a
229
00:37:23,033 --> 00:37:34,133
major source of information; it’s just a big effort to do it that way. But if that were done in a few more, for example, clinical situations where
230
00:37:34,133 --> 00:37:39,166
maybe by luck you can just get enough material and you can section through a glomerulus, we could have a way of saying how bad or how
231
00:37:39,166 --> 00:37:48,932
good Weibel is. What we need in Epon-embedded material cut at very thin sections is some idea of what’s the nuclear shape variation and what’s
232
00:37:48,933 --> 00:37:58,466
the nuclear size variation so we can check some of this to see if Weibel really…are we in a range where it’s very insensitive to changes in this? Is
233
00:37:58,466 --> 00:38:10,332
the normal range of podocyte nuclei shape and size such that it functionally can be ignored? We saw from Kath’s work that very thin sections like
234
00:38:10,333 --> 00:38:17,133
EM, Weibel basically looks like disector-fractionator and what we know from other cases, if you use big thick 4 micron sections,
235
00:38:17,133 --> 00:38:29,899
Weibel is going to fall apart a bit; where is the transition? EM is a big lot of work but if you can cut semis at 1 or .5 microns, if we can find the
236
00:38:29,900 --> 00:38:40,433
right stains, WT1 for example, is it possible we can get something that you can do on light which is going to be a lot easier that gets away from the
237
00:38:40,433 --> 00:38:48,633
thickness effect of Weibel? And then, are there any alternative single-section measurements—that I think we may even hear about today—that
238
00:38:48,633 --> 00:39:01,233
correlate well with true values and can replace even this estimate, kind of like the in vivo things we were talking about, trying to replace doing the
239
00:39:01,233 --> 00:39:09,966
autopsy work? Can we somehow look at the data we get from a population under study, for example the relationship between measured N(A)
240
00:39:09,966 --> 00:39:18,832
and tuft cross-sectional area, and reconstruct something about the radial distribution that allows us to correct Weibel for that and bring it into
241
00:39:18,833 --> 00:39:26,733
alignment? Is it possible that we can use that data in any given group? We would be making assumptions that all patients in that group kind of
242
00:39:26,733 --> 00:39:40,933
behave the same, the distribution was the same, the nuclei were the same, but can we use that to basically fix Weibel? One small twist. We’ve
243
00:39:40,933 --> 00:39:52,933
looked at, in our study of donors, ultrastructure on glomeruli in a large number to look to see if we lose podocytes with age; that was kind of my
244
00:39:52,933 --> 00:40:03,566
suspicion. This is what we found, again based on Weibel-Gomez but as you know now I’m somewhat unapologetic about using it in EM,
245
00:40:03,566 --> 00:40:17,432
though. From 15 to 76 the estimated podocyte…so this is the actual N now we’ve taken from average tuft volume and podocyte density,
246
00:40:17,433 --> 00:40:31,366
shows no sign of going down; 642 in the young, 761, but not significantly different. Why is this the case? Is it that we don’t lose podocytes? We
247
00:40:31,366 --> 00:40:40,366
know from looking at the urine that we do lose podocytes in urine. Now, we do in recent time have a suspicion we may get replacement
248
00:40:40,366 --> 00:40:51,599
podocytes but where are we making these measurements? We’re making them in patent glomeruli. We do not count podocytes in sclerotic
249
00:40:51,600 --> 00:40:57,933
glomeruli for some strange reason. So if, in fact, when you’ve lost enough, and our work in IgA would suggest about a third of the podocytes in a
250
00:40:57,933 --> 00:41:11,233
glomerular tuft it becomes unstable and goes to sclerosis, you may just be censoring out all your abnormal. So, this is a problem with a
251
00:41:11,233 --> 00:41:19,366
cross-sectional analysis. You might say this proves, yes, you don’t lose glomeruli or don’t net-lose glomeruli with age but that’s not necessarily
252
00:41:19,366 --> 00:41:31,966
the case. In patent glomeruli you don’t see the number go down in healthy donor-like people. The problem is, that doesn’t tell you what’s happening
253
00:41:31,966 --> 00:41:40,732
in a normal glomeruli. So even if you had a perfect gold standard measure, even if you did complete enumeration on this, you have to be
254
00:41:40,733 --> 00:41:50,433
careful about what you conclude and I think very similar things were said yesterday in terms of glomerular number. So it’s not always a matter of
255
00:41:50,433 --> 00:41:58,533
doing the right technique and the technique’s limitations, we also have limitations to the material we study and this is, I think, what people were
256
00:41:58,533 --> 00:42:08,466
getting at when they were saying we needed an in vivo reproducible method we can use to look at glomerular number because just looking cross-
257
00:42:08,466 --> 00:42:35,732
sectionally you may not be able to make the right conclusions. So with that, I will close and I’ll be glad to entertain any questions.
258
00:42:35,733 --> 00:42:45,166
FEMALE: Thanks, Kevin, for a wonderful presentation. My question is, much of the studies that we look at in terms of determining shape
259
00:42:45,166 --> 00:42:55,499
factor and things really are done in normal animals or normal humans and I wonder how confident are we that podocyte nuclei do not
260
00:42:55,500 --> 00:43:07,566
change significantly in shape or size in different chronic kidney diseases, so that we can comfortably utilize these equations and
261
00:43:07,566 --> 00:43:13,432
measurement methodology in our clinical samples?
262
00:43:13,433 --> 00:43:24,866
KEVIN LEMLEY: I’m actually only familiar with two studies that estimated shape factor in podocytes, podocyte nuclei. One was by Tim Meyer and that
263
00:43:24,866 --> 00:43:33,332
was looking at Epon-embedded material and looking in profiles, cross-section area, and trying to reproduce an ellipsoid. The other was a
264
00:43:33,333 --> 00:43:43,299
Steffes and that actually worked backwards. That looked at: what’s the disector-fractionator value, what’s the Weibel thing, what beta do you
265
00:43:43,300 --> 00:43:53,000
need to get that? So that didn’t actually measure it, and again, if the radial distribution was the main cause of Weibel giving something wrong, they
266
00:43:53,000 --> 00:44:02,066
may have fit the wrong thing. So, I only know of one place…and I don’t think that’s a good way to do it. I mean, it was a logical way that you look
267
00:44:02,066 --> 00:44:09,866
and you kind of model to your profiles, but someone who’s serially sectioning, I don’t know of a single case where it’s been measured. So [
268
00:44:09,866 --> 00:44:19,999
---] I guess we don’t know in any cases what the actual is, but that’s what I mean by empirically. We need to look at how this…again, it could be
269
00:44:20,000 --> 00:44:30,566
we can a fair amount of variation if it’s in a range because it gets more sensitive the farther off you get in terms of an ellipsoid, the kind of the more
270
00:44:30,566 --> 00:44:39,999
distorted it gets. But I mean, if you just think of what podocyte nuclei look like in EM cross-section, they’re not regular but it’s not like they
271
00:44:40,000 --> 00:44:49,800
look hugely elongated. So my guess is yes, it needs to be investigated, it needs to be investigated in different animal situations,
272
00:44:49,800 --> 00:44:59,900
different human situations, but unless there is a pretty big deviation from spherical, it may not really matter. But until one’s measured it, you can’t
273
00:44:59,900 --> 00:45:06,100
tell. ROGER WIGGINS: Roger Wiggins, Ann Arbor,
274
00:45:06,100 --> 00:45:17,733
Michigan. That was great, Kevin. I am really impressed with the fact that you can eliminate beta—that sounds like a great step forward—and
275
00:45:17,733 --> 00:45:29,866
begin to simplify things. So, can you tell us roughly how much time it would take per sample to do the analysis which you’ve described in
276
00:45:29,866 --> 00:45:32,966
terms of… KEVIN LEMLEY: If you do the simplified analysis
277
00:45:32,966 --> 00:45:42,466
in terms of doing it on EM…so, as I think John was getting into yesterday when he was talking about it, this is after doing this, and after doing
278
00:45:42,466 --> 00:45:50,632
this, and after doing this. So forgetting what the EM tech gets, what my lab person does looking at EM, the A(A) part is a lot harder than the N(A)
279
00:45:50,633 --> 00:46:00,366
part. So I think if you get rid of the A(A), if you use a simplified Weibel, it’s less; it’s still a lot of time. That’s why I think our hope comes from:
280
00:46:00,366 --> 00:46:12,832
can we do something in light in Epon-embedded or resin-embedded sections, that gets away from the thickness problem but that’s a lot easier?
281
00:46:12,833 --> 00:46:20,866
Again, as you well know with something like a WT1 stain, and that’s an empirical thing, is what do you get? Hopefully, if someone wants to take
282
00:46:20,866 --> 00:46:29,599
a lot of time with a gold standard saying, you know, “How thick do I have to get my sections to get that?” So, it’s a long time. When we switched
283
00:46:29,600 --> 00:46:40,433
to doing 5 glomerular profile per donor in Epon we were getting 10 patients done a year and that was kind of in someone’s side time, but it’s very
284
00:46:40,433 --> 00:46:49,733
time-consuming and if you get a lot of good data out of it, it’s worth doing but it’s not a practical method for a lot of labs because it’s very
285
00:46:49,733 --> 00:46:55,466
time-consuming. BEHZAD NAJAFIAN: Kevin, I just wanted to
286
00:46:55,466 --> 00:47:04,466
emphasize something that you also mentioned, that one problem with the model-based method is that you can never get through the assumptions.
287
00:47:04,466 --> 00:47:13,432
So, if you even deal with some of the assumptions that they may not be practically significant, you cannot be comfortable that those
288
00:47:13,433 --> 00:47:28,133
assumptions may hold true or close to true in other conditions, especially in pathology conditions. Another thing would be that here the
289
00:47:28,133 --> 00:47:37,299
data that you presented about N because Weibel-Gomez gives us N(V), right? So, volume of glomeruli you estimated using Weibel-Gomez also,
290
00:47:37,300 --> 00:47:40,933
or…? KEVIN LEMLEY: Yes. It’s not Cavalieri or anything
291
00:47:40,933 --> 00:47:43,866
like that. Everything is Weibel in one way or another.
292
00:47:43,866 --> 00:47:54,732
BEHZAD NAJAFIAN: Everything is Weibel. Yeah. So, do you want to make any comment about estimating numerical density of podocytes using
293
00:47:54,733 --> 00:47:59,466
disector method or no? KEVIN LEMLEY: I’m sorry?
294
00:47:59,466 --> 00:48:06,199
MALE: Do you want to make any comment about it? Because in practice, in the biopsy material, well, you can do Weibel-Gomez method
295
00:48:06,200 --> 00:48:14,800
sequence sections and you can also do numerical density of podocytes using disector method, correct?
296
00:48:14,800 --> 00:48:18,366
KEVIN LEMLEY: Not from single-sections, no. MALE: No, not from single-sections, absolutely,
297
00:48:18,366 --> 00:48:23,166
but that is something that is practically doable. KEVIN LEMLEY: I guess that would be the
298
00:48:23,166 --> 00:48:29,466
question—is it practically doable?—because I don’t see many clinical studies coming out using…
299
00:48:29,466 --> 00:48:32,699
BEHZAD NAJAFIAN: Absolutely not because… KEVIN LEMLEY: I mean, Kath and Rudy did a big
300
00:48:32,700 --> 00:48:40,633
study but that was kind of a research study of technique, so they were more or less obliged to do that. But I would just say that it’s not that you
301
00:48:40,633 --> 00:48:46,666
can’t do a disector there, but when you’re talking doing a disector-fractionator in clinical, I bet you’re making it
302
00:48:46,666 --> 00:48:48,266
a lot less likely it’s going to get done.
303
00:48:48,266 --> 00:48:52,132
BEHZAD NAJAFIAN: Yeah, I agree, but going from numerical
304
00:48:52,133 --> 00:49:03,133
density to number it’s a problem that both Weibel-Gomez and disector-fractionator methods share in terms of impracticality in clinical biopsies.
305
00:49:03,133 --> 00:49:06,866
KEVIN LEMLEY: Just disector; disector-fractionator has number, right?
306
00:49:06,866 --> 00:49:11,299
BEHZAD NAJAFIAN: Yeah, but I mean you can’t do that. It’s very difficult to that in clinical study.
307
00:49:11,300 --> 00:49:17,566
KEVIN LEMLEY: Well I think, just with regard to the different clinical situations, this is where we’re suffering from a dearth. We have John’s
308
00:49:17,566 --> 00:49:23,199
study in mouse where you had a gold standard and you looked at two methods and, in fact, Weibel on an average was off but in some ways
309
00:49:23,200 --> 00:49:34,100
looked like a better predictor of the thing because of the slope and the R-squared. Kath’s study in both controls and Type I diabetes, Weibel at EM
310
00:49:34,100 --> 00:49:43,166
level compared to disector-fractionator basically gave the same values. So, in two different circumstances…I mean, this is the extent of it, the
311
00:49:43,166 --> 00:49:50,566
comparisons. It’s not like we’ve made a lot of comparison, so I think one could say, yeah, in a different system it might be different. We’ve
312
00:49:50,566 --> 00:50:00,566
looked at it in two human systems and if you do Weibel the right way, it gives you the same value. So, I’d say it’s kind of more the impetus is on
313
00:50:00,566 --> 00:50:07,132
people who say Weibel doesn’t work to give me an example where you’ve compared it to a gold standard or to disector-fractionator done right
314
00:50:07,133 --> 00:50:17,433
and it doesn’t agree. Theoretically, lots of things are possible but I think here we just have something that’s a more practical version until it’s
315
00:50:17,433 --> 00:50:25,866
replaced by something else, hopefully, that’s even easier because at EM level Weibel…I think it’s easier, maybe, than disector-fractionator in some
316
00:50:25,866 --> 00:50:34,232
sense. It isn’t like it’s an easy method because EM is just a lot of work.
317
00:50:34,233 --> 00:50:44,299
KATHRYN WHITE: Hi. Kath White, Newcastle. Thank you, Kevin, for a lovely talk. Can I just ask…you showed early on that N(A) correlates
318
00:50:44,300 --> 00:50:47,233
very well with N(V), the podocyte number. Does this hold for other cell types, do you think?
319
00:50:47,233 --> 00:50:52,833
KEVIN LEMLEY: I only looked at it because what I’ve been doing over the years is podocyte estimate, so it was in lots of different
320
00:50:52,833 --> 00:50:56,066
circumstances, but that was always for podocytes.
321
00:50:56,066 --> 00:51:02,832
KATHRYN WHITE: I think sometimes the danger is that people see that that method’s being used and then they say, “Oh, I’ll do it for counting mice
322
00:51:02,833 --> 00:51:11,833
cells,” and it’s not being tested and before you know it people have gone off and just counting something completely different in a different
323
00:51:11,833 --> 00:51:19,799
organ altogether and I think that sometimes we have to be careful of this, that something that’s being shown at one level isn’t taken by other
324
00:51:19,800 --> 00:51:26,900
people and so far down the line it no longer fits and I think that’s maybe what has happened generally with estimating number in things; that
325
00:51:26,900 --> 00:51:35,033
people have taken simplified methods that did fit one system but no longer fit—not in this case but in other cases—and that’s why we’ve gotten to
326
00:51:35,033 --> 00:51:41,033
the point where we’re having to reassess what method’s to be used.
327
00:51:41,033 --> 00:51:47,666
ROBERT CHEVALIER: I agree completely. Bob Chevalier in Virginia. A lot of these talks over yesterday and what you just ended up with,
328
00:51:47,666 --> 00:51:55,599
showing the distribution in the nuclei in a progressive kidney disease where there was drop-out of nephrons and you were actually
329
00:51:55,600 --> 00:52:04,533
looking at the remaining ones is typical of clinical and animal models of progression, and so at some point you have a bimodal distribution; some
330
00:52:04,533 --> 00:52:13,333
glomeruli getting smaller and disappearing and others getting bigger, and of course this has been known for decades, but how is it possible to
331
00:52:13,333 --> 00:52:24,233
mathematically recognize the evolution of this bimodality over time so that what you’re basically looking at was a biased population but due to the
332
00:52:24,233 --> 00:52:29,199
natural history of the process, not due to the technique. How can you recognize that?
333
00:52:29,200 --> 00:52:36,466
KEVIN LEMLEY: Well, the last slide was really a warning that even if technique issues were all solved
334
00:52:36,466 --> 00:52:43,166
you do have the problem of the population you’re studying, and in fact I didn’t make anything of it, but you probably noticed that as the donors got
335
00:52:43,166 --> 00:52:52,466
older the spread of estimated podocyte number increased, and so that may be a reflection very much like Wendy’s looking and John’s looking at
336
00:52:52,466 --> 00:53:01,899
interglomerular volume, that as things get worse and I mean, I don’t consider age something worse, but normal aging may be a place where
337
00:53:01,900 --> 00:53:08,800
you get more heterogeneity. I mean, I don’t want to make too much of the data but I think the problem is, yes, population and what’s happening
338
00:53:08,800 --> 00:53:18,300
on the population level is very pivotal and if you’re going to try to sample that…we’ve found, actually, very low intra-individual variants in
339
00:53:18,300 --> 00:53:26,366
estimated podocyte number; so about one-fourth of what you get between individuals. But if you’re going to get more variants in older donors and
340
00:53:26,366 --> 00:53:35,832
you’re interested in aging, you’re going to have to do more glomerular cross-sections, and the time goes up very badly with that. So I think that’s the
341
00:53:35,833 --> 00:53:43,866
point of that last slide, that even if technique’s not an issue, the population you’re studying is not a trivial thing to deal with and increased variants in
342
00:53:43,866 --> 00:53:48,899
it could kill you in terms of seeing anything from a small sample.
343
00:53:48,900 --> 00:53:55,000
ROBERT CHEVALIER: But we don’t really understand how nephrons disappear and it looks like sclerotic glomeruli don’t stay there forever.
344
00:53:55,000 --> 00:54:01,433
KEVIN LEMLEY: Sclerotic…well, see, that’s a question. Who could look at a sclerotic? Probably only Ichikawa could look at a sclerotic glomerulus
345
00:54:01,433 --> 00:54:08,799
more than once, but the fact is the atubular, I think, and that was a bit of a discussion yesterday, may be the elephant in the room
346
00:54:08,800 --> 00:54:15,266
because I think they do stay there and I agree with Mike; you can recognize them. The Bowman’s capsules are thick, they have
347
00:54:15,266 --> 00:54:23,766
completely patent lumina, they’re very small homogeneous, you know, so you can tell. But, to prove it’s atubular? A lot of work, but the
348
00:54:23,766 --> 00:54:33,299
Marcussen, Ruth Rasch, Tim Meyer, a lot of people have shown in experimental and in human material they may not represent—of
349
00:54:33,300 --> 00:54:41,066
disease—they may not represent a very small part of the population, and Mike made, I think, a good point. They and the compensation that
350
00:54:41,066 --> 00:54:55,066
comes from the loss of their function may be part of what’s behind this spreading out of IgV in normal populations with occult disease, I guess.
351
00:54:55,066 --> 00:55:02,966
ROBERT CHEVALIER: Thank you. MICHAEL MAUER: I’ll just make a comment about
352
00:55:02,966 --> 00:55:16,699
podocyte counting and disease. I remember very well sitting around the table with the whole Ruth Osterby kind of montage when we were looking
353
00:55:16,700 --> 00:55:29,366
at trends. So we have a glomerulus that was [---] in diameter in someone with established diabetic nephropathy and a group of people including
354
00:55:29,366 --> 00:55:36,232
John and myself and other people who had looked at a lot of glomeruli trying to make decisions about what’s a podocyte. Now, that
355
00:55:36,233 --> 00:55:48,066
was easy in the donor biopsies. In disease, it’s difficult, and I think the problem is that the distortion of the tissue and your loss of the sense
356
00:55:48,066 --> 00:56:03,566
of architecture and where we are tends to work towards a reduction. If we were making mistakes it was calling a podocyte something else. So I
357
00:56:03,566 --> 00:56:16,466
think that when disease advances, light microscopy is very, very difficult. I think we know disease changes the markers that we use
358
00:56:16,466 --> 00:56:30,799
histochemically and by light microscopy disease makes counting a real problem. So I think the issues you put forward are correct. Now, how to
359
00:56:30,800 --> 00:56:44,333
get around them? One possibility is that we invest the time and effort and energy for any given condition that we’re looking at to do the gold
360
00:56:44,333 --> 00:56:59,733
standard and then look to how close we can get with other approaches that are more efficient and less time-consuming, but we would need to do
361
00:56:59,733 --> 00:57:12,033
that differently with IgA, differently with diabetes, differently with different diseases that may have different effects on what causes the bias. I’m not
362
00:57:12,033 --> 00:57:22,299
so worried about the time and energy but I think we can put huge amounts of time and energy and money into things that give us wrong answers.
363
00:57:22,300 --> 00:57:33,000
So, I think it’s so important here that we get it right; that that should be our gold standard for accepting papers in this area. At least some of
364
00:57:33,000 --> 00:57:44,500
the work has been done in an absolutely excellent way and then the rest of the work compared to that, so we have a sense of how
365
00:57:44,500 --> 00:57:57,433
off we could be from the gold standard. I don’t think we need to do it in 100 samples, I think we need to do it enough to get a sense of the power
366
00:57:57,433 --> 00:58:02,999
of the less precise method. KEVIN LEMLEY: Well, John Basgen will never be
367
00:58:03,000 --> 00:58:15,633
able to retire, then, if we have to do that…truly retire. I will agree because I’m not familiar with Type I diabetics but of the types of things I’ve
368
00:58:15,633 --> 00:58:22,933
done ultrastructures have been Type II Pima Indians, and unlike IgA where I don’t think there’s usually a question, because of the glomerular
369
00:58:22,933 --> 00:58:34,966
hypertrophy and the breakdown of structure, it can be a real debate what in a good EM section even is a podocyte. So again, that’s separate
370
00:58:34,966 --> 00:58:43,066
from the actual technique you use, that’s just recognition, but the fact that we should always ground them in a gold standard, at least for some
371
00:58:43,066 --> 00:58:47,866
I couldn’t agree with more. I just don’t want to do the gold standard.
372
00:58:47,866 --> 00:58:53,832
KATHRYN WHITE: I’d just like to make one more comment about how we’re sort of saying that doing EM to count podocyte number is very
373
00:58:53,833 --> 00:59:02,733
time-consuming, which it is, but it’s rarely that that is the only thing you’re looking at. You don’t set out to take tissue to just count podocyte number. I
374
00:59:02,733 --> 00:59:10,966
mean, when I did podocyte number in the Type I and often Type II diabetic patients, this was going back to material that had already done the usual
375
00:59:10,966 --> 00:59:17,932
mesangial volume, so it actually took me a very, very short time because all of the images were already there. So, it’s rarely that you would do an
376
00:59:17,933 --> 00:59:25,733
EM study just to look at podocyte number. So if you take in everything you can do on electronmicroscopy, it’s not quite as bad as it first
377
00:59:25,733 --> 00:59:34,133
might appear. ROBERT BACALLAO: Lovely talk. Bob Bacallao.
378
00:59:34,133 --> 00:59:37,433
What happens if you have multi-nucleated podocytes?
379
00:59:37,433 --> 00:59:45,666
KEVIN LEMLEY: Well, Michio Nagata has demonstrated them well and we actually see them in the urine so we know that that may be a
380
00:59:45,666 --> 00:59:55,899
part of them falling off but he, I think, found them in six or seven percent of biopsies that went to a pathologist. If you’re doing it at EM level where
381
00:59:55,900 --> 01:00:07,333
you can really see the cytoplasm, you’re going to say that’s one podocyte and adjust, so exclude the second. But you know, they are kind of
382
01:00:07,333 --> 01:00:16,266
lobulated nuclei and so that’s why he actually did serial sections through to establish that what you’re seeing is two profiles represents a
383
01:00:16,266 --> 01:00:31,066
binucleate podocyte rather than just a bent nucleus seen twice. It’s not quantitatively a huge amount but it does exist in pathologic material.
384
01:00:31,066 --> 01:00:35,766
ROBERT BACALLAO: And nuclear size and shape as regard to disease processes? Can those be…?
385
01:00:35,766 --> 01:00:43,399
KEVIN LEMLEY: Well, I don’t know if we’re going to hear from anybody today on that but there are some…one could raise a lot of theoretical
386
01:00:43,400 --> 01:00:59,066
questions. To me, other than going in to nuclear synthesis, to think of a nucleus really of changing a huge amount in size, unless it’s infected by a
387
01:00:59,066 --> 01:01:07,666
virus or something, again, I don’t know the data where someone’s looked at it in podocytes specifically but my guess is it’s probably not a
388
01:01:07,666 --> 01:01:15,232
huge amount of variation and again, if it’s in a range where the sensitivity of the model—if you’re going to use the model—is low to it, it may
389
01:01:15,233 --> 01:01:19,533
not matter too much. ROBERT BACALLAO: And just as a bit of a
390
01:01:19,533 --> 01:01:26,833
comment, I’ve actually looked at cells when I’ve foamed paraformaldehyde or formalin onto them in culture. Watching the nucleus is a really
391
01:01:26,833 --> 01:01:36,499
instructive event under those fixation conditions. It’s probably stochastic what it does to the nucleus so it won’t come out in your analysis, but
392
01:01:36,500 --> 01:01:46,466
for some pathological processes where people are making statements about where the nucleus is relative to the rest of the cell, they may be
393
01:01:46,466 --> 01:01:50,766
looking at an artifact. KEVIN LEMLEY: Well again, it’s amazing how
394
01:01:50,766 --> 01:02:00,032
much we’re dealing with that hasn’t been grounded in paraffin measurements and maybe that’s not a big surprise in biologic science.
395
01:02:00,033 --> 01:02:09,466
ROBERT STAR: Beautiful talk. I think you’re clarifying certainly for some of us how to get around or how to understand all of the
396
01:02:09,466 --> 01:02:17,099
assumptions that go into this and to figure out which ones are amenable to attack, but I guess the question I have is in thinking about this, is not
397
01:02:17,100 --> 01:02:24,066
so much in the individual glomerulus being able to make that measurement but when you come to the kidney and you ask, “What’s the disease
398
01:02:24,066 --> 01:02:32,199
process and how is the disease process impacting the kidney?” don’t you really want to know distribution of the number of podocytes per
399
01:02:32,200 --> 01:02:41,333
glomerulus across a range of glomeruli, some healthy, some diseased? Isn’t it that distribution that’s important, not the individual number per
400
01:02:41,333 --> 01:02:45,033
glomeruli? KEVIN LEMLEY: Well, there’s a number of strata,
401
01:02:45,033 --> 01:02:52,399
unfortunately, and you’d ultimately like to know everything at every stratum and so the fact is for those of us who do glomerular physiology or
402
01:02:52,400 --> 01:02:59,400
used to when that existed as a field, fundamentally you would like to know a lot within an individual about how different things can be
403
01:02:59,400 --> 01:03:08,766
and correlate it, probably, with like what we used to be able to with rats’ functional measurements in individual nephrons. I think, again, it’s going to
404
01:03:08,766 --> 01:03:15,266
be very dependent. If you do something like the disector-fractionator you are going to have, for the number, you’re going to have it for one
405
01:03:15,266 --> 01:03:26,166
glomerulus. You can reconstruct from that your population. When you combine N(V) N(V) you’re going to be getting, at best for an individual, their
406
01:03:26,166 --> 01:03:34,732
average and sometimes if you need to combine enough things, for example, if you’re going to try to correct for the distribution across a population,
407
01:03:34,733 --> 01:03:42,666
then you’re measuring something at a population level. So we tend to mix things in and out. Ultimately you’d like to have the knowledge of
408
01:03:42,666 --> 01:03:53,666
everything from each individual glomerulus. It just leaves you with a more limited number of techniques you can use to get that. Last
409
01:03:53,666 --> 01:03:59,266
question, I am told. MICHAEL MAUER: Just in response to what Dr.
410
01:03:59,266 --> 01:04:12,166
Star was saying. If we look at foot process width in three glomeruli we find an excellent correlation with urinary protein excretion. So we know that
411
01:04:12,166 --> 01:04:22,399
what we’re looking at is somehow sampling and reflecting the total kidney. Otherwise, by studying three glomeruli, you couldn’t get an association
412
01:04:22,400 --> 01:04:43,366
with total kidney function. If we look at a podocyte parameter at baseline and say that it predicts outcomes, predicts your GFR in 10
413
01:04:43,366 --> 01:05:04,166
years, predicts whether you’re going to live or die, then that becomes an important parameter. In that way we’d have to get some clinical
414
01:05:04,166 --> 01:05:20,832
grounding of the measurement and show that it has relevance to function and outcome, that we get more than just a pathobiologic glimpse, we get
415
01:05:20,833 --> 01:05:32,399
to the question of how much do we have to measure, how carefully, before the estimate that we’re getting at in fact is related to overall
416
01:05:32,400 --> 01:05:44,400
function of the kidney or somehow is a marker of what’s going to happen over time. If we’re looking at it simply as a consequence and people with
417
01:05:44,400 --> 01:05:52,466
diabetic nephropathy have fewer podocytes and
418
01:05:52,466 --> 01:05:58,899
we cannot sort out horse and cart, it’s of less value. So I think the answer to the question is: what is the clinical and functional relevance of
419
01:05:58,900 --> 01:06:06,100
the measure? And the less that we attach it to those, it’s less value.
420
01:06:06,100 --> 01:06:13,166
KEVIN LEMLEY: And only the one example I know of that Rob Nelson and Tim Meyer looked at…I don’t remember if it’s podocyte number or
421
01:06:13,166 --> 01:06:23,032
podocyte density predicting future albuminuria, so that’s about the only example that I know of other than, yes, as you get worse you lose podocytes
422
01:06:23,033 --> 01:06:27,633
of where it has been able to be predicted.
Date Last Updated: 10/3/2012
|
|
