An In-Depth Look at the Faustman Lab Research – Part 2

Dr. Faustman’s lab recently published results from a human trial that began enrolling patients back in 2009 and that was a culmination of many years of research from Faustman’s lab and others around the role of a protein called Tumor Necrosis Factor alpha (TNF-a) in diabetes. According to the publication, a vaccine used against tuberculosis for almost a century, the Bacillus Calmette-Guerin (BCG) vaccine, showed promise as an immune-modifying treatment for long-term diabetes.

In Part 1, we laid out the reasoning behind the study, and went over the biological background. Now we will look at the results reported directly.

Part 2:

The study

In Part 1, we established that the tuberculosis vaccine, Bacillus Calmette-Guerin (BCG), was shown to have preventative effects in mouse models of type 1 diabetes. In a controversial stance, Faustman holds that BCG vaccination’s primary role in this case is to induce TNF-a expression, and that TNF-a’s primary role in this case is to kill the defective autoreactive T cells after they have developed [12], implying that treatment with BCG will only work with patients who are already diabetic.

Faustman put her theory to the test and treated three long-term diabetics with BCG to see whether it would have any effect on their disease status. Now, if you’re paying close attention, you should be stopping me and asking: three? Really? Yes; in the trial there were three diabetic patients treated, and three diabetic patients given a placebo. Three subjects in the real treatment arm seems like precious few, but, given the difficulty and cost of getting human subject trials approved by the FDA, three is a good place to start for a proof-of-concept trial.

The six patients received injections of either a low dose of BCG or placebo twice, in two intradermal injections four weeks apart. For each of the six patients, blood was drawn weekly for eight weeks, followed by bi-weekly draws for four weeks, and a final checkup twenty weeks after the first injection. The placement of patients into treatment groups was randomized, and subsequent measurements were double-blinded. So we’re good on those counts.

Further, Faustman’s group attempted to set up quite a few references for the sake of comparing biometrics from their small treatment groups. Six non-diabetic subjects were paired with the treated subjects. These non-diabetics received neither placebo nor BCG, but had blood drawn at the same time. A total of 73 other individuals were recruited for blood tests to act as references: 41 diabetics were assayed two to five times over the course of the 20 week trial for C-peptide levels; and 16 diabetics plus 16 non-diabetics were used as references for insulin autoreactive T cell measurements (more on what those are later). (In the Methods section of the paper, the author refers to 57 diabetics and another 17 diabetics, but no non-diabetic controls. I am not sure why there is this apparent discrepancy.)

With the six treated patients and the 79 references, Faustman aimed to follow four primary endpoints to determine the efficacy of the BCG treatment.

1. The number of live and dead T cells that respond autoreactively to insulin.
2. The number of regulatory T cells detectable in the blood samples.
3. Antibodies against glutamic acid decarboxylase (GAD), a classical marker of autoimmune diabetes.
4. C-peptide levels, the gold standard of improved pancreatic function in treatment trials.

I will address each of these in turn below. Before I get to the results in each of these endpoints, though, there is one big, honking caveat I have to address first.

A Monkey Wrench

Faustman opens her results section with an explanation: one of the three placebo-treated diabetics, as it turns out, had results that looked much more like the treatment arm than the researchers would have hoped. Based on their analysis of the endpoints, the placebo-treated patient seemed to be responding just like the BCG treated patients.

What had happened? Upon review of the patient’s blood samples, the researchers found that the patient in question had an acute, undiagnosed Epstein-Barr Virus (EBV) infection. Had the researchers or the patient himself known that, the patient would have been excluded from the trial. However, since they only found out during analysis afterwards, the researchers were presented with a quandary. This particular patient’s result was not what was expected, and made it look like the placebo worked as well as the treatment. Instead of three treated and three placebo patients, the researchers had three treated patients, two placebo patients, and a mystery.

The first problem here is the obvious one: they lost one of their controls, dropping the n of this trial from six to five. The second problem here, and the more concerning one, is that Faustman’s group decided in the analysis to treat the one EBV-infected patient as a separate arm, grouped with the BCG arm: “EBV infections, like BCG, trigger innate immunity by inducing secretion of host TNF. The patient’s EBV status and receipt of placebo saline injections fortuitously enabled us to compare the serial T cell and pancreas effects of EBV- and BCG-triggered innate immune responses in the same study.”

On a personal and emotional level, I understand this reasoning. I can empathize the poor researcher upon finding out the control patient had an EBV infections, seeing years of work potentially disappear into nothing. It’s like a long legal trial where you find out at the end that one cop entered without a search warrant, so the whole case falls apart.

However, from an experimental perspective, that is bad. Bad bad bad. The whole point of trials like this is to minimize the number of variables in order to determine the effects of BCG and BCG alone. You can’t just reclassify a control as a treated patient because you decided after the fact that his infection was similar in theory to the treatment you administered. It would be one thing if the patient had tuberculosis. But EBV? That’s a totally different thing; it is a virus, not a bacteria, and, as noted above, the immune system is complex and reacts to different pathogens in different ways. And, the cherry on top, EBV, like many viruses, has evolved techniques to try to evade the human immune system. One of these techniques is to reduce the amount of and the body’s ability to respond to TNF-a [21, 22, 23]. So, if Faustman’s hypothesis is correct and BCG is acting through TNF-a, then EBV might be expected to have an opposing effect. If the virus has the same effect as BCG, we can at best conclude nothing, and at worst conclude that the original theory is wrong.

We will put that aside, though, and accept for now the paper’s claim that the analysis should be conducted as if the EBV patient were just another treatment arm. Moving on to the actual results, then.

Endpoint 1: Insulin Autoreactive T Cells

Any normal individual has some number of T cells that react incorrectly against the body’s own proteins, but in autoimmune diseases like diabetes, these autoreactive T cells are not contained, and they proliferate and launch an all-out attack against self. In the case of type 1 diabetes, insulin is one of the proteins that is identified as “foreign” by some T cells [24]. The T cells that specifically react to insulin circulate throughout the body, albeit at very low levels, and can be used to distinguish new-onset diabetic from non-diabetic blood [25]. These autoreactive T cells seem to persist even in long-term diabetics, at least in the lymph nodes [26].

These autoreactive T cells are exactly the subset that Faustman argues are being selectively killed by TNF-a. In order to test the efficacy of BCG at inducing this selective T cell death, Faustman’s group sets up an assay to measure the number of live and dead T cells that react to human insulin. To do this, they extract T cells from the blood samples of each patient, and then expose the T cells to a particular segment of the insulin protein held in a way an autoreactive T cell would recognize and bind to a fluorescent molecule. The T cells that recognize that complex hold on to the fluorescent molecule, which can then be measured using flow cytometry. Flow cytometry is a way of measuring the properties of groups of cells using lasers, and a flow cytometer can measure the size, granularity, and fluorescent characteristics of many cells at once [27]. So the autoreactive T cells, now bound to a fluorescent molecule, can be counted using a flow-cytometer.

So far, so good; the theory behind this assay is reasonable. However, to see why I take issue with the results presented, it is important to understand that flow cytometry is a bit of an art form. You don’t feed in cells and get a single number printed out; you get a map of the spectral properties of the cells that were analyzed, and it is the job of the scientist to determine what is noise and what is true signal. To understand what I mean, take a look at Supplementary Figure S2 from the paper.

Here, Faustman is showing how dead and live cells were counted. On the left-hand side, you see the output of the flow cytometer in red. Each red dot is a cell, graphed according to its forward scatter (a measure of size) and its side scatter (a measure of granularity). These two properties are often used to distinguish different populations of cells, and can also be used to give a rough estimation of live versus dead cells, since dead cells tend to shrink up and become more granular. Here, one of the researchers has drawn “gates” around what he or she sees as two populations of cells and added labels. These gates are not automatic or obvious; drawing them correctly is a matter of experience. Ideally, we would want to see a big margin between two populations. However, as you can see in the left-hand image, what the researcher has called dead and alive is all mashed together. There are also many red dots left outside the gates; these the researcher presumably has determined to be debris or other contaminating cells, neither dead nor alive. This, again, is a matter of decision on the part of the researcher, and each individual would come up with a slightly different result.

How accurate is this separation of live and dead cells by size and granularity? In order to determine how accurate the two gates are, the Faustman group also did a test with propidium iodide (PI) stain that is presented on the right-hand side. PI adds fluorescent molecules only to dead cells, so it is a more precise measure of dead versus live cells than size and granularity. You can see that with this more accurate measure, only 92.4% of the cells that the researcher called dead in the left-hand image were actually dead, while 7.60% were actually alive. This, then, gives us a measure of error that says that about 7% of cells the researchers are calling “dead” are actually alive. And, to make matters more complicated, the lines dividing the PI stain images are also drawn by a researcher using the same principles– so he could move that line up or down depending on his best judgment, resulting in a higher or lower percentage of misidentified cells.

In other words, any percentage presented from these images can be tweaked based on human judgment. This is important, because the researchers use this gating of live versus dead cells based on size and granularity to determine how many live and dead autoreactive T cells each patient has. So before we even get to the actual measurements, we know we might have a 7% error rate in our labeling of dead cells.

(I know, flow cytometry is kind of wild– lasers? Stick with me here.)

With that background, then, we can look at what Faustman is actually reporting: according to this assay, within one to four weeks of the BCG treatment, there was an increase in insulin autoreactive T cells in the blood of treated patients that was not seen in the placebo-treated diabetic patients or in the paired non-diabetic controls (those who received no treatment but had blood drawn for the sake of comparison). Figure 5 from the paper below shows the kind of output from the flow cytometer that was used to determine these results. As before, the lines that are drawn are drawn by an individual, based on judgment and experience, and the separation of the live versus dead cells is based on the gating discussed above, with an error rate of around 7%. Anything above the horizontal line is thought to fluoresce sufficiently to be called insulin autoreactive, and it is not clear to me from the figure or the text what the x-axis is, but perhaps forward scatter, which measures size. The percentage reported is the percentage of red dots that appear in the upper-right quadrant in each image, which is supposed to represent insulin autoreactive T cells.

Imagine moving the horizontal line up or down a tick; all the percentages change, some more than others. So why is the horizontal divider exactly where it is? That’s where the analyst put it. Without direct access to that person, I presume he drew it approximately where the large block of cells at the bottom seemed to end in one of the control samples, and then used that gate for all the subsequent samples. But, clearly, there is some error and wiggle-room in that line; after all, the control sample, which the caption says should have no dead or live autoreactive T cells, shows up to 0.4% insulin autoreactive T cells based on these images. So now we have up to 0.4% error in identifying autoreactivity, plus some 7% of the total cells might be misidentified as either dead or alive. And we see that in this particular control, the number of cells identified as autoreactive changes from week to week, so we know there is some intra-sample variability here.

Then, looking at the percentages assigned to the treated samples, we see that we’re talking very small margins of change. The percent of live insulin autoreactive cells ranges from 0% to 0.3%; that is within the margin of error established by the control sample, so we can’t conclude anything. The percentage of dead autoreactive cells ranges from 0.1% to 1.0%, but we know that 7% of those cells might be alive, and there’s maybe 0.4% of error based on the control samples, and if you nudge that dividing line up a hair as might be warranted by the researcher’s judgment, you lose a  lot of those allegedly dead insulin autoreactive cells. So what are we left with? Maybe something is happening there; I can’t say that certainly nothing is happening. But with such large margins of error relative to the true signal, I don’t think we can safely conclude anything.

Okay, now let’s look at Figure 4; perhaps that will clarify things. Here, we see the same data about insulin autoreactive T cells, but reduced to a graph and presented for all of the patients. At first glance, it looks convincing; the purple controls seem so steady as compared to the wildly fluctuating treated patients.

However, there are several very concerning parts of this figure:

1. This is derived from the same flow cytometric analysis, so all the fuzziness described above still applies.
2. The authors state that the non-diabetic references had maximally 0.4% insulin autoreactive cells showing up, which they say is error above 0%. However, in these graphs, the controls seem to hold steadily around 0.5%. The minor difference is not important, but the discrepancy is confusing.
3. Look at the graphs in A.v. These are sampled from the two pairs of 16 reference subjects, which the authors perform the same assay with to see what “normal” diabetic insulin autoreactive percentages look like. The authors say there are two groups– either patients have no autoreactive T cells apparent (as in image A.iv.), or they consistently have autoreactive T cells present (as in image A.v.). However, for most of these patients, there are only two or three data points; the patients appear to be segmented into groups based on whether they have more insulin autoreactive T cells showing up than the paired non-diabetic control. However, as detailed above, there is variability and error in these measurements, so it is unclear how the researchers could confidently set up these two classes of patients– those that have insulin autoreactive T cells and those that don’t– based on so few data points.
4. As a corollary to [3], the researchers state that before treatment, all the diabetic patients receiving BCG were of the type that had no insulin autoreactive T cells. They concluded this based on a single data point before treatment began, as far as the figures show. Given the aforementioned error and variability, the researchers would need to establish the nonexistence of insulin autoreactive T cells over several measurements; one measurement alone could just be a chance low value.
5. The graphs in A.iv. and A.v. betray a potential problem with this assay. Notice how the reference diabetic and the non-diabetic in each graph tend to have parallel movement? If one goes up, so does the other, so that the shape of the two lines is similar? This could very well just be random similarity that makes a pretty pattern. Or it could be that the researchers were hand-pairing subjects for these graphs, which means we could re-pair the references and perhaps reclassify some of the “no insulin autoreactive T cells” group as having more insulin autoreactive T cells than a particularly low control. Or it could be what is called a “batch effect,” where samples from each batch have similar technical artifacts. Imagine, for example, that each pair has their blood drawn at the same time by the same person, and each pair is analyzed at the same time. Each pair, then, might end up with slight bits noise that come from the handling process, unique to each batch.
6. In both A and B, the BCG-treated subjects show significant peaking of insulin autoreactive T cell measurements. That looks promising in terms of the claims of the paper. However, note the little black arrows above the graphs. These represent the two BCG injections. If the BCG were causing the spikes in insulin autoreactive T cells, the injections would precede any spikes. However, what we see in several cases that the spikes start, then the first injection occurs. Further, after the second injection, it’s unclear what we should expect, but it’s strange to see that some patients see an increase in measured insulin autoreactive T cells, and some see a decrease. What could possibly explain the fact that the increase precedes the first injection, and that the second injection yields entirely inconsistent results?

The details of these assays, then, are questionable. More concerning still is that there are several crucial controls missing. Let’s assume for a minute that the claims are accurate, and that we are seeing an increase in insulin autoreactive T cells in the treated patients, and especially dead T cells. Faustman’s hypothesis says that this is a result of TNF-a selectively killing off the autoreactive T cells. An alternative explanation is that the patient has just been injected with live bacteria, leading to a system-wide immune response that in turn leads to the proliferation of certain T cells and the death of other T cells, non-specifically, as the body readjusts its defenses to fight off the current enemy. In order to cross off this alternate explanation, Faustman would need to show that only autoreactive T cell numbers are increasing in the patient blood. However, we see no measurements of other specific T cells or even T cells as a whole. The Methods section mentions that T cells reactive to EBV were also assayed for all samples, but that data is not presented anywhere. That could be because the data was uninteresting– those cells did not show any change in number, so why include it? Or it could be because the data was unsupportive– those cells did show a change in number, implying the alternate explanation is plausible. My inclination is to assume the former, but, in either case, the data is necessary to rule out the alternate explanation.

What, then, can we conclude from the first of the four endpoint assays? For a skeptic like me, at worst we can say that there is too much noise in the data to conclude anything at all, and at best we can say that the system-wide infection has caused some T cell death. We cannot say, though, that TNF-a is selectively killing off insulin autoreactive T cells.

Endpoint 2: Regulatory T Cells

Regulatory T cells (Tregs) are a special population of T cells that suppress the activity of other T cells during immune response, thereby helping prevent autoimmunity. Treg levels in the peripheral blood have been linked to the body’s ability to avoid autoimmunity, with decreased levels of Tregs observed in certain autoimmune diseases [28]. The Faustman lab reasons, therefore, that an increase of levels of Tregs in peripheral blood samples from treated patients would imply that the body is increasing its regulatory activity, suppressing the T cells that lead to autoimmunity.

To begin with, this is a somewhat simplistic picture of how the regulatory system operates; though low levels of Tregs are associated with autoimmunity, there are also autoimmune diseases in which patients have higher levels of Tregs, and autoimmune diseases, type 1 diabetes being one, where Tregs in the circulating blood don’t seem to reflect what’s happening in the affected organ [28]. Further, regulation is about more than just numbers; some studies indicate that dysfunctional interactions between Tregs and T cells contribute to autoimmunity in diabetes [28]; if that is the case, then an increase in circulating Tregs would not necessarily indicate a decrease in autoimmunity.

With that caveat in place, let’s look at the data. The researchers use a similar flow-cytometry assay as for insulin autoreactive T cells to measure the number of Tregs in each patient sample, this time using fluorescent molecules that bind to proteins specific to the regulatory T cell population. This endpoint is given one brief paragraph in the paper because, as you can see in figure 6A below, the results are inconclusive. Two treated patients and the EBV infected patient show increases in Treg counts, but one treated patient shows no change versus the paired non-diabetic control. So, in two out of three patients we see an increase in Tregs, which may or may not indicate a change in immune regulatory behavior systemically, and says nothing about immune regulatory behavior around beta cells.

Even assuming that we see some increase in Tregs in the peripheral blood, this endpoint presents yet another problem: generally speaking, it is not surprising for a systemic bacterial infection to increase circulating levels of Tregs [29]. As the T cells of the immune system proliferate to respond to the attacker, the regulatory T cells also proliferate to keep the T cells in check. At best, then, monitoring Tregs in the peripheral blood confirms that the BCG injection worked, and that the patient’s immune system is responding to an influx of bacteria. In this context, monitoring Tregs says nothing concrete, unfortunately, about the patient’s autoimmune tendencies.

Endpoint 3: GAD Antibodies

One of the predictors of diabetes is that B cells begin to make antibodies against some of the body’s own proteins, including glutamic acid decarboxylase (GAD), an enzyme found in islets in the pancreas [30]. Measuring the levels of GAD autoantibodies, therefore, is a predictor of diabetes, and many long-term diabetics continue to produce GAD autoantibodies [31].

Because GAD antibodies are such a strong indicator of type 1 diabetes, a reversal of autoimmunity would conceivably come with a reduction in GAD antibody levels. Faustman’s group therefore monitored antibody levels in the six treated diabetics throughout the course of the study. As with the Treg measurements, however, the results were variable; one treated patient showed no GAD antibodies at all, one treated patient showed an overall increase; and one treated patient showed an overall decrease. Looking at figure 6B, we can see that the two placebo-treated patients also showed variation in antibody levels, with one patient moving from 650 units to somewhere around 450 units, a decrease of approximately 30%. The authors of the paper say none of the placebo-treated patients showed significant changes in antibody levels, but that 30% is more than the 10-20% change in the BCG-treated patient that they claim significantly changed antibody levels, so the reasoning there is unclear.

More importantly, though, any changes are inconsistent; down in one patient, up in another, and nowhere in the third. The researchers also assay several other diabetes-associated autoantibodies, and get similarly variable results, uncorrelated to the times of injection or the treatment arm.

Once again, then, we are left with inconclusive evidence. Were GAD autoantibody levels affected by the treatment? Maybe, but this data does not indicate confidently in either direction.

Endpoint 4: C-peptide Levels

And so we arrive at our final endpoint. C-peptide is a fragment of the pro-insulin protein, the precursor to the insulin protein that beta cells release. C-peptide gets released from beta cells like insulin, and is thought generally to be a byproduct of the insulin production process. Measuring C-peptide is the standard way to measure the body’s ability to produce insulin, since diabetics who do not produce any insulin have no C-peptide even when they have injected insulin. Thus, C-peptide measurements are the gold-standard primary endpoint for diabetes treatment clinical trials; increase C-peptide levels, and we can say beta cells are producing more insulin.

In this preliminary study, Faustman was not expecting to see huge increases in C-peptide levels, or to see that the beta cells had regained their functionality. The researchers were instead hoping to see any change at all that would indicate that maybe even a handful of beta cells were regenerating and beginning to produce insulin. Given the low levels of C-peptide the researchers were interested in observing, the standard C-peptide assays were not sensitive enough; the Roche Cobas C-peptide assay used to select patients for the trial, for example, has a lower threshold of 330 – 470 picomols per liter, which is far below the level of a functional pancreas, but too insensitive if what we’re interested in is any change at all.

So, for the C-peptide assays, the Faustman group sent the blood samples to Mercodia AB in Sweden, which makes an ultrasensitive C-peptide kit that can measure C-peptide values between 1.5 and 285 pmol/L [32]. Despite the fact that none of the selected patients were positive for C-peptide in the initial, less sensitive assay, all six had detectable levels of C-peptide using the ultrasensitive assay. (The fact that long-term diabetics still produce measurable C-peptide is not news [32], but it is still really cool. Every time I hear that, part of me jumps for joy. There is hope! If we can just re-balance the immune system, the little beta cells might stand a fighting chance!)

With a minimal but detectable baseline of C-peptide, the researchers looked at what happened after administration of BCG or placebo. With two of three BCG treatments and the EBV infection, the authors state, there is a transient but significant increase in C-peptide levels. Does this hold up on closer inspection?

Looking at figure 8A and B, the increase that the researchers see is not so clear. Perhaps there are spikes in patient ii and the EBV infected patient; but patients i and iii look fairly similar to the two placebo patients. The authors are saying that patient i shows an overall increase, but, if we’re allowing transient increases, what exactly separates the placebo group from the treated group? If you removed the labels from the six images, I certainly would not be able to place them back in their correct categories.

More importantly, though, we can assume that the transient increases are real, and we still have a problem: there is no way to tell from this data whether an increase is a specific reduction of autoreactive T cells, or just a generic response to systemic infection. There are a number of well-documented cases in which autoimmune diseases like Multiple Sclerosis and Graves disease seem to go into “remission” during pregnancy, as the mother suppresses her own immune system in order to allow a foreign being to grow inside of her unperturbed [33]. Clinical trials currently ongoing are investigating whether a similar phenomenon might exist in type 1 diabetes. If such transient remissions of autoimmunity occur in response to physiological changes like pregnancy, it is possible that similar remissions might occur during systemic infection; I do not know of any published data on this, but I imagine with the availability of ultrasensitive C-peptide assays, it is only a matter of time. This is just speculation, but perhaps systemic infection “distracts” the immune system, allowing for transient increases in beta cell mass and thus insulin production that dissipate as soon as the infection passes. That would be interesting scientifically, and would potentially offer insight into new therapies, but would not be equivalent to saying BCG treatment is causing a specific relapse of autoimmunity.

In other words, the C-peptide levels look shaky at best, and it is possible that any transient increases are informative about the body’s response to infection without indicating specific depletion of autoreactive cells.

Conclusions

And so we’ve reached the end of the paper. All four endpoints, upon close inspection, are interesting but inconclusive. What does this mean– should we throw out this line of research and pillory Faustman? Certainly not. Contrariwise, I think the role of the innate immune system in type 1 diabetes is an area begging for more attention, and Faustman does a good job of bringing some attention to it. However, this particular study lacks a degree of rigor and consistency that we will need in both molecular and clinical experiments if we want to find an answer that really works.

Why do I care so much? Why so much nit-picking over a single study? Well, to be honest, I care because other people care; Dr. Faustman, more than any other researcher in this space, seems to have a fan base that follows her every word. That is fine in theory, but we as a diabetic community should be pushing forward the best research and the best researchers so that we can reach a cure sooner, and so it pains me to see mediocre research lauded by the patient community.

I believe a cure is possible in my lifetime, and our scientists will need our support to do the research that will bring it to us. The answer is not to back the scientist who promises the biggest payoff, but rather to insist on sufficient funding for well-reasoned yet ambitious projects. If you’re not sure which projects those are, start by asking the JDRF [link]; they have made it their business to review and evaluate diabetes science. If you think the JDRF has become unfocused on a cure, then I suggest organizations like the Pediatric Diabetes Research Center or the Diabetes Research Institute.

On the other hand, if you are like me and think that if you want something done you have to do it yourself, then get to reading, and start asking questions, and don’t stop until you’ve found the cure– I look forward to thanking you for it one day!

[1] Bradley JR. TNF-mediated inflammatory disease. J Pathol. 2008 Jan;214(2):149-60. http://www.ncbi.nlm.nih.gov/pubmed/18161752

[2] Nelson DE, Ihekwaba AE, Elliott M, Johnson JR, Gibney CA, Foreman BE, Nelson G, See V, Horton CA, Spiller DG, Edwards SW, McDowell HP, Unitt JF, Sullivan E, Grimley R, Benson N, Broomhead D, Kell DB, White MR. Oscillations in NF-kappaB signaling control the dynamics of gene expression. Science. 2004 Oct 22;306(5696):704-8. http://www.ncbi.nlm.nih.gov/pubmed/15499023

[3] Coppieters KT, Dotta F, Amirian N, Campbell PD, Kay TW, Atkinson MA, Roep BO, von Herrath MG. Demonstration of islet-autoreactive CD8 T cells in insulitic lesions from recent onset and long-term type 1 diabetes patients. J Exp Med. 2012 Jan 16;209(1):51-60. http://www.ncbi.nlm.nih.gov/pubmed/22213807

[4] Lehuen A, Diana J, Zaccone P, Cooke A. Immune cell crosstalk in type 1 diabetes. Nat Rev Immunol. 2010 Jul;10(7):501-13. http://www.ncbi.nlm.nih.gov/pubmed/20577267

[5] Schulte BM, Kramer M, Ansems M, Lanke KH, van Doremalen N, Piganelli JD, Bottino R, Trucco M, Galama JM, Adema GJ, van Kuppeveld FJ. Phagocytosis of enterovirus-infected pancreatic beta-cells triggers innate immune responses in human dendritic cells. Diabetes. 2010 May;59(5):1182-91. http://www.ncbi.nlm.nih.gov/pubmed/20071599

[6] Dahlén E, Dawe K, Ohlsson L, Hedlund G. Dendritic cells and macrophages are the first and major producers of TNF-alpha in pancreatic islets in the nonobese diabetic mouse. J Immunol. 1998 Apr 1;160(7):3585-93. http://www.ncbi.nlm.nih.gov/pubmed/9531322

[7] Yang XD, Tisch R, Singer SM, Cao ZA, Liblau RS, Schreiber RD, McDevitt HO. Effect of tumor necrosis factor alpha on insulin-dependent diabetes mellitus in NOD mice. I. The early development of autoimmunity and the diabetogenic process. J Exp Med. 1994 Sep 1;180(3):995-1004. http://www.ncbi.nlm.nih.gov/pubmed/8064245

[8] Jacob CO, Aiso S, Schreiber RD, McDevitt HO. Monoclonal anti-tumor necrosis factor antibody renders non-obese diabetic mice hypersensitive to irradiation and enhances insulitis development. Int Immunol. 1992 May;4(5):611-4. http://www.ncbi.nlm.nih.gov/pubmed/1510787

[9] Christen U, Wolfe T, Möhrle U, Hughes AC, Rodrigo E, Green EA, Flavell RA, von Herrath MG. A dual role for TNF-alpha in type 1 diabetes: islet-specific expression abrogates the ongoing autoimmune process when induced late but not early during pathogenesis. J Immunol. 2001 Jun 15;166(12):7023-32. http://www.ncbi.nlm.nih.gov/pubmed/11390446

[10] Hayashi T, Faustman D. Defective function of the proteasome in autoimmunity: involvement of impaired NF-kappaB activation. Diabetes Technol Ther. 2000 Autumn;2(3):415-28. http://www.ncbi.nlm.nih.gov/pubmed/11467344

[11] Kodama S, Davis M, Faustman DL. The therapeutic potential of tumor necrosis factor for autoimmune disease: a mechanistically based hypothesis. Cell Mol Life Sci. 2005 Aug;62(16):1850-62. http://www.ncbi.nlm.nih.gov/pubmed/15968469

[12] p2 in the paper in question

[13] http://en.wikipedia.org/wiki/Bacillus_Calmette–Guérin

[14] Mollenkopf HJ, Hahnke K, Kaufmann SH. Transcriptional responses in mouse lungs induced by vaccination with Mycobacterium bovis BCG and infection with Mycobacterium tuberculosis. Microbes Infect. 2006 Jan;8(1):136-44. http://www.ncbi.nlm.nih.gov/pubmed/16257251

[15] Qin HY, Sadelain MW, Hitchon C, Lauzon J, Singh B. Complete Freund’s adjuvant-induced T cells prevent the development and adoptive transfer of diabetes in nonobese diabetic mice. J Immunol. 1993 Mar 1;150(5):2072-80. http://www.ncbi.nlm.nih.gov/pubmed/8436836

[16] Yagi H, Matsumoto M, Kishimoto Y, Makino S, Harada M. Possible mechanism of the preventive effect of BCG against diabetes mellitus in NOD mouse. II. Suppression of pathogenesis by macrophage transfer from BCG-vaccinated mice. Cell Immunol. 1991 Nov;138(1):142-9. http://www.ncbi.nlm.nih.gov/pubmed/1833073

[17] Lee IF, van den Elzen P, Tan R, Priatel JJ. NKT cells are required for complete Freund’s adjuvant-mediated protection from autoimmune diabetes. J Immunol. 2011 Sep 15;187(6):2898-904. http://www.ncbi.nlm.nih.gov/pubmed/21844383

[18] Shehadeh N, Calcinaro F, Bradley BJ, Bruchim I, Vardi P, Lafferty KJ. Effect of adjuvant therapy on development of diabetes in mouse and man. Lancet. 1994 Mar 19;343(8899):706-7. Erratum in: Lancet. 2006 Apr 15;367(9518):1240. http://www.ncbi.nlm.nih.gov/pubmed/7907682

[19] Dahlquist G, Gothefors L. The cumulative incidence of childhood diabetes mellitus in Sweden unaffected by BCG-vaccination. Diabetologia. 1995 Jul;38(7):873-4. http://www.ncbi.nlm.nih.gov/pubmed/7556994

[20] Allen HF, Klingensmith GJ, Jensen P, Simoes E, Hayward A, Chase HP. Effect of Bacillus Calmette-Guerin vaccination on new-onset type 1 diabetes. A randomized clinical study. Diabetes Care. 1999 Oct;22(10):1703-7. http://www.ncbi.nlm.nih.gov/pubmed/10526739

[21] Gosselin J, Menezes J, D’Addario M, Hiscott J, Flamand L, Lamoureux G, Oth D. Inhibition of tumor necrosis factor-alpha transcription by Epstein-Barr virus. Eur J Immunol. 1991 Jan;21(1):203-8. http://www.ncbi.nlm.nih.gov/pubmed/1846816

[22] Gosselin J, Flamand L, D’Addario M, Hiscott J, Menezes J. Infection of peripheral blood mononuclear cells by herpes simplex and Epstein-Barr viruses. Differential induction of interleukin 6 and tumor necrosis factor-alpha. J Clin Invest. 1992 Jun;89(6):1849-56. http://www.ncbi.nlm.nih.gov/pubmed/1318324

[23] Morrison TE, Mauser A, Klingelhutz A, Kenney SC. Epstein-Barr virus immediate-early protein BZLF1 inhibits tumor necrosis factor alpha-induced signaling and apoptosis by downregulating tumor necrosis factor receptor 1. J Virol. 2004 Jan;78(1):544-9. http://www.ncbi.nlm.nih.gov/pubmed/14671137

[24] Haskins K, Cooke A. CD4 T cells and their antigens in the pathogenesis of autoimmune diabetes. Curr Opin Immunol. 2011 Dec;23(6):739-45. http://www.ncbi.nlm.nih.gov/pubmed/21917439

[25] Seyfert-Margolis V, Gisler TD, Asare AL, Wang RS, Dosch HM, Brooks-Worrell B, Eisenbarth GS, Palmer JP, Greenbaum CJ, Gitelman SE, Nepom GT, Bluestone JA, Herold KC. Analysis of T-cell assays to measure autoimmune responses in subjects with type 1 diabetes: results of a blinded controlled study. Diabetes. 2006 Sep;55(9):2588-94. http://www.ncbi.nlm.nih.gov/pubmed/16936208

[26] Kent SC, Chen Y, Bregoli L, Clemmings SM, Kenyon NS, Ricordi C, Hering BJ, Hafler DA. Expanded T cells from pancreatic lymph nodes of type 1 diabetic subjects recognize an insulin epitope. Nature. 2005 May 12;435(7039):224-8. http://www.ncbi.nlm.nih.gov/pubmed/15889096

[27] http://en.wikipedia.org/wiki/Flow_cytometry

[28] Long SA, Buckner JH. CD4+FOXP3+ T regulatory cells in human autoimmunity: more than a numbers game. J Immunol. 2011 Sep 1;187(5):2061-6. http://www.ncbi.nlm.nih.gov/pubmed/21856944

[29] Rowe JH, Ertelt JM, Way SS. Foxp3(+) regulatory T cells, immune stimulation and host defence against infection. Immunology. 2012 May;136(1):1-10. http://www.ncbi.nlm.nih.gov/pubmed/22211994

[30] Schlosser M, Walschus U, Klöting I, Walther R. Determination of glutamic acid decarboxylase (GAD65) in pancreatic islets and its in vitro and in vivo degradation kinetics in serum using a highly sensitive enzyme immunoassay. Dis Markers. 2008;24(3):191-8. http://www.ncbi.nlm.nih.gov/pubmed/18334741

[31] Falorni A, Brozzetti A. Diabetes-related antibodies in adult diabetic patients. Best Pract Res Clin Endocrinol Metab. 2005 Mar;19(1):119-33. http://www.ncbi.nlm.nih.gov/pubmed/15826926

[32] Greenbaum CJ. Dead or alive? Diabetes Care. 2012 Mar;35(3):459-60. http://www.ncbi.nlm.nih.gov/pubmed/22355015

Karmel Allison

Karmel was born in Southern California, diagnosed with Type 1 Diabetes at the age of nine, and educated at UC Berkeley. Karmel now lives in San Diego with her husband, where she is loving the sunshine, working in computational biology at the University of California, San Diego, and learning to use the active voice when talking about her diabetes.

Karmel Allison

Karmel was born in Southern California, diagnosed with Type 1 Diabetes at the age of nine, and educated at UC Berkeley. Karmel now lives in San Diego with her husband, where she is loving the sunshine, working in computational biology at the University of California, San Diego, and learning to use the active voice when talking about her diabetes.