Curcumin Trial Results: Antiviral Effect Reported

The first human trial of curcumin -- an "alternative"
treatment in use in the AIDS community, and the first of a
potentially new class of anti-HIV treatments which target the
LTR (long terminal repeat) of the AIDS virus -- found a
modest antiviral effect, according to SEARCH Alliance, a
community-based research organization in Los Angeles. This
was only a pilot study, however; and the effect appeared to
be temporary, as it had partially faded by 20 weeks when this
study ended. More research is needed -- such as the new
curcumin trial, about to begin in Boston, which will be
conducted by the Community Research Initiative of New
England.

The antiviral effect was seen in measurements of HIV RNA,
using an experimental test (quantitative PCR) which is very
sensitive, but also highly sensitive to laboratory errors and
variations. P24 antigen tests, which are much less sensitive,
failed to confirm an antiviral effect; also, there was no
improvement in average T-helper count. The only toxicity seen
at the dose tested (which is the most popular dose in use in
the community), was upper gastrointestinal discomfort in some
of the volunteers; it was severe enough to cause two of the
19 to drop out of the trial.

Curcumin is found in the spice turmeric; it is the substance
which gives curry its yellow color. It first came to the
attention of the AIDS community after laboratory studies at
the Dana-Farber Cancer Institute, and follow-up viral tests
at Beth Israel Hospital in Boston, showed anti-HIV activity
in the laboratory. Despite publication of these laboratory
findings more than a year ago in the Proceedings of the
National Academy of Sciences, USA (March 1993; volume 90,
pages 1839-1842), no company or government agency is studying
curcumin as a possible treatment, nor funding anyone else to
do so. SEARCH Alliance, which specializes in fast-track
development of the earliest pilot studies of potential AIDS
treatments, found private funding for its small trial, and
reported the results on May 3.

Curcumin is the weakest known agent in its class (of
substances found to inhibit the HIV LTR); it was chosen for
the trial because more powerful LTR inhibitors were not
available for this research. All information suggests that
this treatment technology could be developed much further. We
hope the SEARCH Alliance results will bring more attention to
a very important research area which has been neglected by
the mainstream. But it is also important to remember that the
results so far are limited and sometimes ambiguous; they do
not yet prove that curcumin is beneficial.

[Note: This writer is a member of the medical/scientific
advisory board of SEARCH Alliance; however, this article
represents only our views, not necessarily those of the
organization. We originally suggested a curcumin trial to
SEARCH Alliance; however, we were not involved in designing
or conducting the study, and we first saw the data on May 2,
a week before this issue went to press.

[This article is based on "Curcumin As an Antiretroviral in
HIV-1 Positive Patients," an unpublished manuscript by Robert
E. Winters, M.D., Michael J. Slattery, Charles V. Chesson II,
Ph.D., and Natalie L. Sanders, M.D. -- and on additional
information provided by SEARCH Alliance. Our analysis differs
from that of the SEARCH team -- we used median values,
instead of mean values, to summarize the data, as explained
below -- but while the numbers are different, our conclusions
are the same.]

Notes on Interpretation of the Data

This was a small pilot study with a total of 19 patients
enrolled, and 11 completing the trial. There was no placebo
or other control group. As a result, this trial will not
provide hard conclusions; it cannot prove that one treatment
is better than another. What it can do is to give us early,
suggestive information about a new kind of potential
treatment in human use -- a more complete and accurate view
than can be obtained from collecting anecdotal reports.

Therefore, this article will not focus on conclusions, but
rather walk through the key data, pointing out what we find
interesting and important.

In both the scientific and activist AIDS communities at this
time, there is a widespread conservative view that trial
results must only be analyzed according to procedures
specified in advance. The purpose of this limitation is to
prevent errors which can occur when researchers look at the
data and then concoct theories to fit it; if dozens of
different theories are considered, some are likely to fit
just by chance alone, when no real discovery has been made.

We believe that this conservative view is correct for certain
large trials which are seeking definitive proof of the value
of a treatment. But it should not apply to initial pilot
studies, which aim to help researchers find out what is going
on, not to test or prove a preconceived idea. All too often,
researchers blindfold themselves, creating theories and
designing trials abstractly, without looking at enough real
data; usually these theories are wrong, and then the
resulting trials are not useful, except as miscellaneous
negative results. The consequence is inefficient,
unproductive research. A better way is to run a pilot study
first, and look at all of its data in various ways, asking
what the data is trying to tell us; then use that information
to design a more formal, definitive trial.

In the curcumin study, a related issue concerns analyzing
subsets of the data. This study sought volunteers with
advanced HIV disease; the median T-helper count at entry was
100. Several participants dropped out of the trial, usually
because of opportunistic infections; and any infection or
other immune activation is likely to cause a substantial rise
in HIV RNA, the principal viral measure used in this study,
making the treatment look less active than it otherwise
would. As a result, when all the patients who entered the
trial are analyzed together, the antiviral effect observed is
small; but when those who completed the trial without an OI
are analyzed -- or those who entered it with a higher than
average T-helper count -- the effect is much more clear. We
present the data both ways.

Also, in this article we have summarized the data using
median values, instead of mean (average) values. The median
of a set of numbers is the 50-percent point -- the value such
that half of the numbers are above it and half are below it.
We decided to use the median instead of the average, because
the most important data (the measurement of HIV RNA) are
highly variable, with a number of extremely high and low
values which may be due to laboratory errors. The median is
often used for looking at ill-behaved data, since a few
extreme values can only cause a limited error in the median,
no matter how high or low those values may be -- while a few
extreme values can completely distort an average.

Study Design Overview

The purpose of the SEARCH Alliance curcumin trial was to look
for any signs of toxicity of the treatment, at the dose
commonly used in the AIDS community -- and also to record any
indications of antiviral activity. For this pilot study, all
patients were enrolled in a single treatment arm; there was
no control group.

To check for toxicity, the study included "a complete blood
count with differential, chemistry panel, amylase, and a
self-evaluation questionnaire at baseline and at the end of
week 1, 2, 4, 6, 8, 12, 16, and 20." (Quotations are from the
SEARCH Alliance manuscript, "Curcumin As an Antiretroviral in
HIV-1 Positive Patients," unless otherwise stated.)

To look for anti-HIV activity, "Viral load was monitored by
serum RNA-PCR at baseline and at week 2, 4, 8, 12, 16, and
20." (The manuscript includes a technical overview of the PCR
test used, and of procedures for quality control.) Also,
three different kinds of p24 antigen tests were used, in an
effort to find one which gave enough positive values at
baseline to allow useful comparisons.

The study was initially designed to last eight weeks.
However, the data at week four was encouraging enough that
the trial was extended to 20 weeks.

All study volunteers received the same dose of curcumin.
"Patients were instructed to take three capsules of
concentrated curcumin, three times a day, one hour before or
two hours after meals. Each capsule, supplied by Nature's
Herbs, weighed 445 mg. and contained approximately 285 mg. of
curcumin concentrate in a ground turmeric base. The total
amount of curcumin ingested per day was 2.56 grams." (The
volunteers were judged "very compliant as determined by
counting the number of returned capsules," meaning that they
actually took the intended study dose.) In case stomach upset
became a problem, the curcumin could be taken with meals.

A total of 19 volunteers entered the study. One dropped out
after two days, however, and therefore was not included in
the analysis of antiretroviral data below, since only a
baseline test was done, so there is no data on changes in
viral tests.

Inclusion/Exclusion Criteria

Study volunteers could have any T-helper count; however, an
effort was made to enroll persons with a low count, so that
they would be more likely to have p24 antigen values which
could be measured. For the 18 patients, the T-helper count at
baseline ranged from 10 to 391, with a median of 100.5.

Of the total of 19 who enrolled, 11 finished all 20 weeks of
the study without an opportunistic infection. (A 12th has 20
weeks of data, but was not included with the other 11 due to
an OI.)

Volunteers could use AZT or other antiretrovirals during the
trial, provided they were on a stable treatment regimen for
60 days preceding enrollment (so that changes in antiviral
treatments, such as starting or stopping AZT or ddI, would
not interfere with the viral measurements in the trial.) All
of the 11 patients who finished the trial were using
antiretrovirals; we do not have information on the others.
Only one started an antiretroviral treatment during the trial
-- patient number 02, who started d4T. Use of vitamins,
minerals, and herbs during the study was strongly
discouraged.

Inclusion criteria also included hemoglobin, white blood
count, granulocytes, platelets, bilirubin, ALT, AST, SGPT,
alkaline phosphatase, and creatinine.

The 19 participants included 15 men and 4 women. There were
two Mexican-Americans, two African-Americans, and 15
Caucasians. Their ages ranged from 29 to 59 years old.

Results

[Note: The data discussed in this article is reproduced in
Table I, below. The charts (Figure I through Figure III)
could not be reproduced here. But the data used to create the
charts -- the median values, and the counts on which those
medians are based -- are shown, below Table I. Even without
the charts, you can use this data to follow the discussion.

[For a copy of the original article including Figure I
through Figure III, send a self-addressed stamped envelope
to: AIDS Treatment News, P.O. Box 411256, San Francisco, CA
94141, or call 415/255-0588.]

One of the 19 volunteers who entered the study, who had a
history of peptic ulcer disease, withdrew after two days due
to gastrointestinal irritability. This person is not included
in the 18 patients analyzed below, since only the baseline
tests were given, so there are no data on changes in HIV RNA.

One other patient also withdrew due to gastrointestinal
irritability, after eight weeks. Four others withdrew due to
opportunistic infections, and two others left for non-medical
reasons. No patient was lost to analysis. Aside from the
gastrointestinal irritability, no indications of toxicity
were found.

The study measured HIV RNA by polymerase chain reaction
(PCR). It also measured p24 antigen, an older test of viral
activity. In addition, T-helper count and percent, CD8 count
and percent, complete blood count with differential, and some
blood-chemistry values, were recorded throughout the trial.

The primary indicator of viral load -- HIV RNA in the blood,
as measured by quantitative PCR -- is presented in the tables
and charts below.

Table 1 shows the raw data on HIV RNA for all of the patients
in this study (except for the one who dropped out after two
days). Also, the baseline T-helper count for each patient is
shown. Even a casual look at this table suggests that those
with high T-helper counts appear to be responding better than
others to the treatment. Note that the charts below (Figure 1
through Figure 3) can all be reproduced from Table I, with no
additional information required.

Figure 1 shows the HIV RNA data graphically. Also, for each
time point, the median RNA value at that time, and the number
of patients in that median, are shown below the chart. Notice
that there is a substantial spread of HIV RNA values at
baseline (week 0) and at week 2. But later (with the possible
exception of week 8, which is discussed separately), the
pattern is that the most of the values are clustered at the
low end of the scale -- often clustered so tightly that the
separate lines cannot be distinguished on the graph. Besides
these low values, there are a few "outliers," very high
values which are entirely different from the rest. We do not
know what these high values represent; some could be
laboratory errors. [Note: In these graphs (Figure 1 through
Figure 3), a point by itself only means that the data points
adjacent to it are missing, so there is no line connecting
them.]

At week 8, the median is high, about where it was at
baseline. But, as one can verify from the data in Table 1, a
change of a single value would have made the median more than
six times lower. The data at week 8 shows a very unusual
distribution, with six values under 1,000, two values which
are both 3900, and the other nine values beginning at 25,000.

If laboratory error accounts for a single one of these high
values at week 8, the medians would show a clear antiviral
effect; even as it is, the medians still suggest an antiviral
effect. And this is with all data included, even from those
patients who had opportunistic infections, which are likely
to greatly raise the level of HIV RNA.

Note that no one dropped out of the study until the week 8
data (there is one missing value at week 4, but this person
did not drop out at that time). Only one participant had
dropped out by week 8; and three more dropped out by week 12.
A low dropout rate is important, because those who are doing
poorly in a trial are the most likely to leave; this creates
a selection bias, which can make a drug look good even if it
really does nothing. With this study, a close look at the
data shows that the dropouts could not have affected the
medians very much, even through week 20, no matter how high
their HIV RNA values might have been.

Figure 2 shows the HIV RNA for patients who completed the
trial without an opportunistic infection. Here the antiviral
effect looks very clear, with the median viral load showing a
drop of 20-fold or more between weeks four and 12. At weeks
16 and 20 the medians have climbed, but they are still
several times lower than when the trial began. This suggests
that the antiviral effect may be temporary, with viral load
rising again by sixteen weeks.

Figure 2 must be viewed with caution, however, because of the
likelihood of selection bias. Those who stayed with the study
to the end (and therefore were counted in Figure 2) would be
likely to be those who were doing well. Therefore, the
results shown in Figure 2 are likely to be too optimistic.

Figure 3 shows that patients with less advanced HIV infection
seem to respond better to the treatment (by having a larger
drop in viral load) than the group as a whole. It shows the
changes in HIV RNA for those patients whose T-helper count at
entry to the study was in the upper half of the group. This
turns out to be those with a T-helper count over 100. The
medians do show a good antiviral response in this group. Each
median is based on a small number of data points; but there
is a clear pattern, with the medians being much higher at
baseline and at week 2 than later.

Figure 3 is important because it has less opportunity for
selection bias than Figure 2. We deliberately selected the
most healthy patients (at enrollment) for Figure 3; this is a
legitimate choice, as the trial could have made the same
selection through its inclusion criteria, and Figure 3 shows
what would have happened if it did. But (except for a few
dropouts near the end of the trial), Figure 3 avoids the
major potential bias of Figure 2 -- the retrospective
elimination of some patients who turned out to do poorly (a
choice which could not possibly have been made at the time
the trial began). The fact that Figure 3 shows a substantial
drop in HIV RNA after week 4 provides reassurance that the
antiviral effect is real.

Figure 3 is also reassuring in that it is hard to imagine any
laboratory bias that could cause those with higher T-helper
counts at study entry to show more improvement than those
with lower counts.

Figure 3 also suggests that future curcumin trials might find
better results in persons with higher T-helper counts.

Discussion and Commentary

The results of this trial could be interpreted differently.
Because of the wide variability in the data, and the lack of
confirmation of antiviral activity by p24 antigen tests, one
could conclude that the possible anti-HIV effect of curcumin
in people has not been shown. The absence of p24 antigen
reduction is surprising, since it was anecdotal reports of
just such an effect, completely unexpected and unlikely to
have been due to chance, that led to the curcumin trial in
the first place. Some uncertainty will remain until other
studies confirm or contradict these results.

Meanwhile, we note that the HIV RNA reduction was seen
despite the fact that a number of circumstances probably made
the response lower than it otherwise would have been.

First, this trial deliberately recruited persons with more
advanced HIV disease, in order to get enough viral load for
the p24 antigen test to measure. The data suggests that those
with less advanced HIV infection might have responded better.
(See discussion on viral suppression, below.)

Also, four of the 18 volunteers in this trial had
opportunistic infections -- which can greatly increase the
HIV viral load, by causing immune activation, which leads to
the growth of HIV.

In addition, ten of the 18 volunteers received their flu
shots during the trial. This probably affected the data; at
least two research groups have reported that the immune
activation caused by flu shots can temporarily raise the
level of HIV RNA. Four patients received flu shots within
three weeks of a measurement of HIV RNA, and all of them
showed large increases -- about 10-fold for two of the four.
This probably made the antiviral response to the treatment
appear less than it really was. (Patient 7 received a flu
shot one week before the blood draw which showed a very high
value at week 8, contributing to the unusual data for week 8,
which was discussed above.)

How do these results compare to what is known about other
antivirals? Definitive figures are hard to obtain, since
testing for HIV RNA is still experimental and fairly new;
there is much variation within each test, and between the
different tests. But it is generally believed that AZT will
produce about a one log reduction (10-fold reduction) of
viral load as measured by HIV RNA.

The best viral suppression so far has been with some of the
new protease inhibitors, which can temporarily produce a
reduction of viral load by two to three logs (about 100-fold
to 1000-fold). New information is suggesting that even
persons with very advanced HIV disease (for example, with a
T-helper count under 10) do show immune recovery and
clinical benefit, if the virus can be suppressed enough. (At
this time, unfortunately, that result can only be achieved
temporarily, since HIV develops resistance to the drugs. This
is why it is so important to test new drug combinations, such
as one or more protease inhibitors together with other
drugs; in other diseases, such as tuberculosis and certain
cancers, combinations of several drugs with different
mechanisms of action have been successful in overcoming major
problems of drug resistance. For the same reason, it is also
important to test LTR inhibitors -- an approach not accepted
at this time, nor even widely understood, in the AIDS
research mainstream. If this kind of treatment works, it
could become an important part of drug combinations.)

It might not matter how the virus is suppressed, as long as
it is kept low enough. This is why it is also critically
important that better tests for viral load or activity become
available for clinical practice. Then physicians will be able
to tell quickly if a treatment regimen is working, or if it
needs to be changed.

The SEARCH Alliance trial fits into this picture by showing
the potential for developing an entirely new class of
treatments -- drugs which target the LTR of HIV, not to kill
the virus but to keep it inactive. Remember that curcumin is
the weakest known agent of this class; it is attractive
because it is widely available and relatively nontoxic, since
it has long been used in food. But it may not be strong
enough to help much by itself, especially when HIV disease
is already advanced. SEARCH Alliance is currently developing
plans to test more powerful LTR inhibitors, such as beta
lapachone, CPT-11, or topotecan.

Better HIV treatments are urgently needed. And a new kind of
treatment would be especially important, by contributing to
drug combinations which attack multiple targets of the virus.
The SEARCH Alliance curcumin trial has provided the first
human data on a new approach to AIDS drug development --
screening for compounds which inhibit the HIV LTR. While much
more research is necessary, this trial has already suggested
that the approach does work in people.

For More Information

For more information, contact SEARCH Alliance, 7461 Beverly
Blvd., Suite 304, Los Angeles, California, 90036, 213/930-
8820.


Table I: HIV RNA Data

Patient# week2 week4 week8 week12 week16 week20 CD4
baseline (baseline)
p01 180 4,000 1 70 5 10 10 344
p02 31,800 120,000 24,000 3,900 40 8,700 137,000 107
p03 86,300 20,800 400 120,000 7,800 27,600 2,700 133
p04 100 450 1 200 80 5 320 181
p05 28,500 13,300 12,400 25,000 530 6,000 3,700 78
p06 31,900 34,600 6,400 3,900 910 3,900 5,800 219
p07 15,000 41,000 14,000 150,000 900 7,000 3,700 15
p08 58,500 12,000 500 960 4,500 2,800 2,600 94
p09 38,400 18,000 200 88 890 9,100 5,000 46
p10 20,600 10,000 500 400 80 8,000 710 116
p11 1,854 61,500 1 270 800 3,000 850 391
p12 55,000 20,000 1 95,600 25,000 68,100 49,000 56
p13 78,400 46,900 16,000 25,200 1,400 88
p14 2,500 15,400 10,000 130,000 8,000 33
p15 13,700 25,000 1,000 83,900 142
p16 21,800 52,100missing 180,000 127
p17 16,000 100,000 11,800 76,100 32
p18 100,000 45,590 25,000 10


** Figure I Data: [Note: Figure I to Figure III not included]

baseline week2 week4 week8 week12 week16 week20
Median 25150 22900 1000 25000 895 6500 3200
Count 18 18 17 17 14 12 12

** Figure II Data:

baseline week2 week4 week8 week12 week16 week20
Median 28500 18000 500 960 800 6000 2700
Count 11 11 11 11 11 11 11

** Figure III Data:

baseline week2 week4 week8 week12 week16 week20
Median 20600 25000 450 3900 80 3900 850
Count 9 9 8 9 7 7 7