1084 GOLDSCHMIDT& WOLF: JOURNAL OF AOAC INTERNATIONAL VOL. 90, No.
4, 2007
FOOD COMPOSITION AND ADDITIVES
Determination of Niacin in Food Materials by Liquid
Chromatography Using Isotope Dilution Mass Spectrometry
ROBERT J. GOLDSCHMIDT and WAYNE R. WOLF
U.S. Department of Agriculture, Agricultural Research Service, Beltsville Human Nutrition Research Center, Food
Composition Laboratory, Beltsville, MD 20705
The availability of deuterium-labeled nicotinic acid
makes stable isotope dilution mass spectromerty
(MS) coupled with liquid chromatography (LC) an
attractive option for the determination of the
water-soluble 13-vitamin niacin in food samples. A
method was developed based on acid digestion,
solid-phase extraction with a strong cation
exchange column, and reversed-phase
chromatography with a C18 column. Detection is
by positive ion electrospray MS. Analysis in
selected ion recording mode is subject to
interference problems similar to those found with
other LC determinations of niacin, but the
additional selectivity of multiple reaction
monitoring mode largely eliminates interference
problems. The method was applied to 6 different
food matrixes and to appropriate reference
materials, including milk samples with niacin levels
near I ppm. The method exhibited good accuracy,
based on levels obtained for the reference
materials, and relative standard deviations in the
range of 0.5-5%.
resent methodology for the measurement of niacin in
foods consists mainly of 3 types: microbiological,
p colorimetric, and liquid chromatography (LC) using
UV detection. AOAC Official Methods are
microbiological (1) or colorimelric (2). Microbiological
methods are sensitive but laborious and time consuming.
Standard levels recommended in the AOAC Methods are in
the range of 10-400 ppb niacin. Reported relative standard
deviations (RSD) are often near 5% (3-6). Colorimetric
methods are faster but lack specificity and are not as sensitive.
Most recent reports on niacin measurement describe the use of
LC with UV detection (3-5, 7-12). There can be significant
interference from food matrixes due to the low specificity at
the LTV region of interest (Xmax = 260 rim for nicotinic acid).
Successful analysis usually requires either significant sample
cleanup prior to chromatographic analysis (3-5, 8, 9, 12, 13)
Received December 29, 2006. Accepted by SG March 26, 2007.
Author to whom correspondence should be addressed; e-mail:
wayne.wolf@ars.usda.gov
or extra chromatographic steps, such as column switching (4).
As interferents may be matrix-specific, published LC-UV
methods may be limited in applicability. Reported limits of
detection (LODs) are generally in the range of
200 ppb—1 ppm. Better sensitivity and specificity can be
achieved with post-column derivatization and fluorescence
detection (6, 14), with reported LODs as low as 5 ppb (14).
Most of the published LC methods have not been rigorously
validated.
LC coupled with detection by mass spectrometry (MS)
offers the potential for excellent sensitivity and specificity.
Sample preparations suitable for niacin analysis by LC-UV
and other methods should also be suitable for LC/MS.
Chromatographic methods developed for LC-UV or
LC-fluorescence methods would in most cases have to be
modified, however, due to issues such as ion suppression and
problems created by involatile species in commonly used
mobile phases. Niacin has a relatively low molecular mass
(123 amu), and thus in selected ion recording (SIR) mode
there may still be significant interference problems. With
instruments that allow multiple reaction monitoring (MRM),
mode however, interference problems can potentially be
avoided, even without optimized chromatography or sample
cleanup.
The method presented here makes use of stable isotope
dilution (ID) MS for quantitative analysis of niacin.
Isotopically labeled versions of both nicotinic acid and
niacinamide are commercially available at a reasonable cost.
The use of an isotopically labeled internal standard has
distinct advantages in quantitative analysis, as it can correct
for analyte losses and makes possible high levels of accuracy
and precision (15). Sample digestion and cleanup are based on
a previously published LC-UV method (8, 9).
Niacin can be found in foods in the forms of nicotinic acid
and niacinamide, and in free or bound forms (16-18).
Fortification of foods may be with either nicotinic acid or
niacinamide, though the latter is more common. Interpretation
of reported niacin values in foods is not always
straightforward, as bound forms of niacin, particularly in
grains, may not be readily bioavailable. Sample preparation
typically involves either acid or alkaline digestion, although in
some cases where only free niacin is of interest, extraction
may be sufficient. Alkaline hydrolysis is usually reported to
give total niacin levels, and acid hydrolysis to give a measure
 GOLDSCHMIDT &WOLF: JOURNAL OF AOAC INTERNATIONAL VOL.
of available niacin. However, the specific hydrolysis
conditions (e.g., acid strength) can affect the completeness of
the measurement. From a nutritional standpoint, tryptophan
levels are also of interest, due to the in vivo conversion of
tryptophan to niacin; however, tryptophan measurement is a
separate issue (16 5 19).
In this work, nicotinic acid was the form monitored, as the
acid digestion method used converts any niacinamide present
to nicotinic acid. The relatively harsh acid digestion method
used here should give a complete measure of niacin for most
food samples; however, in some cases, such as grain sources,
an alkaline digestion may be necessary for measurement of
total niacin.
Experimental
Samples
Reference material (RM) 8437 (hard red spring wheat
flour) and RM 8435 (whole milk powder) were obtained from
the National Institute of Standards and Technology (NIST;
Gaithersburg, MD). Multivitamin tablets [Standard Reference
Material (SRM)] 3280 were also obtained from NIST.
Commercial flour, breakfast cereal, and milk samples were
obtained from local retail markets. Milk sample NFY0409F6
was obtained from the Nutrient Data Laboratory (NDL) of the
Beltsville Human Nutrition Research Center [U.S.
Department of Agriculture (USDA), Beltsville, MD )J, as was
Brand E multivitamin tablet. RM 25C (breakfast cereal) was
obtained from Medallion Laboratories (Minneapolis, MN).
USP Reference Standard nicotinic acid was obtained from
U.S. Pharmacopeia (Rockville, MD). 134-nicotinic acid (99.4
atom % D, labeled at the ring carbons) was obtained from
CDN Isotopes (Quebec, Canada). Standard natural and
labeled niacin solutions were gravimetrically prepared using
0.1M HC1 as solvent. Labeled niacin solutions were
calibrated by reverse isotope dilution against gravimetrically
prepared solutions of the USP Reference Standard niacin, and
all natural standard and labeled solutions were periodically
checked against freshly made solutions in order to monitor
accuracy of the calibrations.
Sample Preparation
Sample preparation was based on AOAC Peer-Verified
Method 1:2001 (PVM 1:2001) for the determination of niacin
in infant formula by solid-phase extraction (SPE) and LC with
UV detection (8, 9). Because the detectability of niacin by the
MS detector is better than that of a UV detector, sample
amounts were, in most cases, smaller than indicated by the
guidelines in PVM 1:2001. The 134-niacin spike was added at
the time of sample weighing. Samples were treated with
1:1 H2SO4, autoclaved, treated with 7.5 M NaOH and then
additional 1:1 H2SO4, and filtered as specified in PVIVI
1:2001. The SPE step specified in PVM 1:2001 was modified
as follows: Aromatic sulfonic acid SPE columns
(6 mL/1000 mg columns from J.T. Baker, Phillipsburg, NJ)
were pretreated with one column volume of methanol
followed by one column volume of 0.1 M, pH 2. 1,
90, No. 4,2007 1085
H 3 PO4/NaH2PO4 buffer. The filtered samples were then
applied to the columns (1-6 mL). Samples were washed with
10-12 mL of the pH 2.1 phosphate buffer, and then niacin was
eluted with 10-12 mL of 0.1 M, pH 5.6, acetic acid—sodium
acetate buffer.
Chromatography and Mass Spectrometry
Most experiments were performed with an Agilent (Palo
Alto, CA) 1100 LC system and a Phenomenex (Torrance, CA)
Luna C 18(2) column (150 x 2.0 mm, 3 tm particle size), fitted
with a Phenomenex SecurityGuard C18 guard cartridge. A
water (A):acetonitrile (B) mobile phase was used (plus
0.1% formic acid). The flow rate was 0.20 mL/min, with a
starting composition of 5% B, a linear gradient to 50% B from
5.0 to 8.0 mm, holding at 50% B to 13.0 mm, a linear gradient
back to 5% B from 13.0 to 16.0 mm, and holding at 5% B to
20.0 mm. The niacin retention time was about 2.5 mm. SIR
experiments were performed using a Beckman (Fullerton,
CA) 114M pump and a Higgins Analytical
(Mountain View, CA) Targa C 18 column (250 x 4.6 mm, 5 pm
particle size). In these cases, the isocratic mobile phase was
80:10:10 25 mm formic acid/ammonium
formate: methanol: acetonitriie with a flow rate of 0.5 mL/min.
The detector was a Waters (Milford, MA) Quattromicro triple
quadrupole mass spectrometer operated in positive
electrospray ionization (ES!) mode. Both natural and labeled
nicotinic acid are protonated in positive ion EST, giving
quasi-molecular ions at mass-to-charge ratios (m/z) of 124 and
128, respectively, which were monitored directly in the SIR
experiments and selected as the parent ions in the MRM
experiments.
Calculations
Calculations of niacin levels proceed from the ratio defined
by:
R areaM! = (Xnat)(Ainai)+(Xsp)(Aisp)
area M2 (X ,rn )(A 2nI ) + (XXsp )(A 2,p
where areas M 1 and M2 are integrated peak areas for the
selected, isotopically related ions (m/z 124 and 128 for SIR, or
124 - 80 and 128 - 84 for MRM, respectively, in this work);
Xna t and X, are the amounts of the natural and isotopic spike
material present in the sample, respectively; and the A terms
refer to the relevant isotopic abundances of the natural and
spike materials. This equation is solved for Xnat , the ratio
being experimentally determined, and all other terms being
known. In the case of natural and 134-niacin, there is no
contribution of the natural to the signal monitored for the
spike, or of the spike to the signal monitored for the natural.
Thus solving for Xnat gives:
Xna t = C*R*Xsp
where R is experimentally determined, X 5 , is known, and C is
a factor introduced to correct for instrument bias. C is obtained
 1086 GOLDSCHMIDT & WOLF: JOURNAL OF AOAC INTERNATIONAL VOL. 90, No.
1.8
1.6
14-
12-
'a
1-
0.8
06
04
0.2
02 0,4 0.6 0.8 I 1,2
1,4
ppm natural niacin (gravimetric)
Figure 1. 124/128 peak area ratio versus ppm natural
niacin for fixed amount (1 ppm) D4-niacin. SIR mode.
by running a standard containing both natural niacin and
labeled niacin. It is then given by:
C = ( Xna t IXsp)( I /R)
where all terms refer to the standard used. All niacin
measurements are thus calibrated to the USP standard.
Results and Discussion
Area Ratio Curves and Detection Limits
Area ratio curves for standard solutions in SIR mode are
shown in Figures 1 and 2, in which the experimentally
determined (m/z 124)I(m/z 128) peak area ratio is plotted
against ppm and ppb unlabeled niacin, respectively. In the first
ease (Figure 1), the D4-niacin level was held constant at
1 ppm, and the natural niacin level was varied from 0.3 to
1.5 ppm using a gravimetrically prepared stock solution of
20.5 ppm natural nicotinic acid. A linear response was
observed, with signal-to-noise (S/N) ratios all over 1000 (rms
noise). In the second case (Figure 2), the D4-niacin level was
held constant at 10 ppb, and the natural niacin level varied
from I to 16 ppb (stock solution of natural nicotinic acid was
0.104 ppm). Again a linear response was observed, but with
greater variance and with S/N ranging from about 5 to
80 (rms). The data shown in Figures 1 and 2 were obtained
using the Higgins column, but similar linear responses were
obtained with the Phenomenex column in both SIR and MRM
modes. Quantitation in stable IDMS is generally not
performed through calibration • curves such as those in
Figures 1 and 2 but is instead determined from individual
runs, as described earlier. As a measure of accuracy and
precision under 'best case", low level conditions, an estimate
of the natural niacin level of the stock solutions used can thus
be obtained for each of the individual runs comprising the data
in Figures 1 and 2 and compared to the known, gravimetric
values. The 10 estimates from Figure 1 yield an average of
20.5 ± 0.2 ppm (RSD = 1.5%), in excellent agreement with the
gravimetric value of 20.5 ppm. The 10 estimates from Figure
2 yield an average of 0.111 + 0.008 ppm (RSD 10%), also in
agreement with the .gravirnetric value of 0.104 ppm. Not
4, 2007
1.8
00 0.8
06 OxO385
2
9894
10 15
1.6
ppb nat niacin (gravimetric)
Figure 2. 124/128 peak area ratio versus ppm natural
niacin for fixed amount (10 ppb) D4-niacin. SIR mode.
surprisingly, precision is worse at the ppb level than at the
ppm level, as interfering signal becomes relatively more
important and baseline definition for peak area determinations
becomes more arbitrary. LOD is estimated as slightly less than
I ppb, about a 200 or more times improvement compared to
values typically reported for LC-UV determinations of niacin.
The SIR mode has the advantage of being straightforward
and is the quantitative mode of choice for instruments not
equipped to select and fragment parent ions. With appropriate
mass spectrometers, however, MRM mode is generally
preferable for quantitation, as it adds an additional level of
selectivity and greatly improves S/N. There is an isotopic
effect observed in the MRM mode analysis of nicotinic acid
by IDMS. The natural nicotinic acid daughter spectrum
(Figure 3b) suggests that the transitions 124-80 or 124-78
would be suitable for MRM analysis. The fragment at m/z 80
corresponds to a loss of CO2 and that at m/z 78 to losses of
H2 0 and CO. For the latter, a corresponding transition in
electron impact (El) is 123—*78 and involves losses of
hydroxyl and carbon monoxide, but it also involves an
exchange between the hydroxylic and 3-hydrogens (20). The
presence of the ring deuteriums in D4-nicotinic acid results in
different relative intensities for the corresponding fragment
peaks (m/z 84 and 81, Figure 3a) compared to those of natural
nicotinic acid. Use of a correction factor in JDMS calculations
for MRM mode (see Experimental) is thus important for
accuracy. In our work with niacin, correction factors used in
MiJv1 mode monitoring the transitions 124—*80 (natural) and
128—*84 (D4) have generally been within 5% of unity.
Table I contains the results of niacin determinations
performed in MRM mode for selected food and dietary
supplement materials, including appropriate reference
materials. For the commercial food materials, expected values
are as listed on the packaging. NIST does not provide a
reference value for niacin for R1'vl 8437; however,
determinations of its niacin value have been reported in the
literature (8, 21). The expected value for RM 25C was
provided by Medallion Laboratories. The expected value
listed for the Brand E multivitamin tablets was supplied by
NDL, but is an approximate value, as described below. SRM
 GOLDSCHMIDT& WOLF: JOURNAL OF AOAC INTERNATIONAL Vol.. 90, No. 4, 2007 1087

841
100-
a) 134-nicotinic acid
12&0
I 81.1
%-

75.8 86 871 -90.99i 91250

01 -

80,1
1001
b) natural nicotinic acid
78,1 123,9

•%.

90

60
635
65
681

70 75
I 80
83.1 - 87.3 , 91.70
85 90 95
983
100
106.0
l032_j_
105
111.91150
110 115
122.0 1256 129.4
120 125 130

Figure 3. Daughter spectra of (a) D4-nicotinic acid (b) natural nicotinic acid and with ionization by positive ESI
and collision gas argon.

3280 does not have an official reference value for niacin as
of this writing. The expected value listed in Table I for SRM
3280 is based on results reported in ref. 22.
Precision obtained for the samples in Table 1, expressed
as RSD, ranged from 0.5 to 2.7%. Mean levels of niacin
obtained for the commercial flour and cereal samples are
higher than the expected values, but this is not surprising, as
such samples are often fortified in excess of the claimed
amounts. Results for the flour (RM 8437) and cereal
(Medallion 25C) RMS are in agreement with the expected
values. Note that the value reported for RM 8437 in Table 1
is on a dry mass basis (measured moisture level 8.9%, following
the procedure from the NIST Report of Investigation for RM
8437). All other values reported in Table I are on an "as is"
basis.
It is difficult to assess the accuracy of the results for the
multivitamin samples without a true reference value. The level
obtained for Multivitamin Brand E is in agreement with the
(approximate) expected value, which corresponds to 100% of
the Daily Allowance for a 2000 calorie diet and is consistent
with the label claims of many commercial brands of daily

Table 1. Determination of niacin in food samples and multivitamin tablets

Sample Niacin, ppm, expected (s)° Niacin, ppm, by LCSlDMSb

537 79.4 ± 2.4 (n = 4)

Brand A wheat flour
Brand B corn meal 40c 68.0 ± 1.1 (n = 4)
Brand C brown rice flour 46c 59.4 ± 2.7 (n = 4)
RM 8437 (HRSW) 39.4 (n = 2)i 40.0 ± 0.3 (n = 4)
40.6 (5.4; n =
161c 279.0 ± 11.9 (n = 4)
Cereal Brand D
Medallion Cereal RM 25C 302.1 (31.1) 267.4 ± 2.9 (n = 4)
12801 ± 522 (n = 4)
Multivitamin Brand E 13000 (20 mg/tablet) (19.81 mg/tablet)
NIST Multivitamin (SRM 3280) (ppm value not given) 16171 ± 375 (n = 4)
(23.01 mg/tablet 0.67)' (24.78 mg/tablet)

8 Values in parentheses (s) are sample standard deviations; n is the number of samples analyzed, where applicable.
Given ranges are 95% confidence limits. SIDMS = stable isotope dilution mass spectrometry.
C
Expected values are label claims on commercial products.
d Reference 21.

Reference 8.
Reference 22.
1088 GOLDSCHMIDT &WOLF: JOURNAL OF AOAC INTERNATIONAL VOL. 90, No. 4, 2007

Table 2. Determination of niacin in milk samples "milk, whole, 3.25% milkfat" and a value of 0.84 ppm niacin
Expected niacin Niacin, ppm, for "milk, producer, fluid, 3.7% milkfat." We thus assume that
Sample level, ppm by LC-SIDMS° expected values for niacin for whole milk samples are near
1 ppm. The levels obtained for the commercial samples are in
RM 8435 (whole milk 7.65 ± 0.93 7.28 ± 0.37 (n = 4) the expected range, and that obtained for RM 8435 is in
powder, dry) agreement with the reference value. The level reported for
Whole milk Brand F 1 ppm 0.911 ± 0.070 (n = 4) RM 8435 is on a dry mass basis (measured moisture level of
2.7%). The value obtained for sample NFY0409F6 is lower
Whole milk Brand C 1 ppm 0.968 ± 0.016 (n 4)
than the others, but is in the range of other milk samples we
NFY0409F6 1 ppm 0.638 (n = 1)
have tested for NDL. Sample NFY0409F6 was also
NFY0409F6 1 ppm 0.657 (by standard determined by a standard additions experiment, using 4 levels
additions)
of added niacin and a constant amount of the D4-niacin spike
Given ranges are 95% confidence limits, n is the number of (Figure 5). The standard additions curve exhibits good
samples analyzed. linearity and has a slope close to unity, consistent with
complete recovery of added niacin and with an absence of
matrix interference. The level obtained from the additions
experiment, 0.657 ppm niacin, is in reasonable agreement
multivitamin tablets. The results for SRM 3280 reported in with the level obtained from the single LC-IDMS
this work are roughly 10-15% higher than those previously determination reported in Table 2.
obtained by this laboratory using an LC/MS method (22).
Work is underway to resolve this discrepancy. Conclusions
Table 2 contains results of niacin determinations of some
milk samples, again in MRM mode, including RM 8435 An LC-IDMS method has been shown to be an appropriate
(whole milk powder) from NIST. These samples are not choice for measurement of niacin in food and dietary
fortified and contain niacin at considerably lower levels than supplement samples. The method described was applied to
the samples in Table I. Precision obtained for these samples is 6 different matrix types, and with appropriate RMS wheat
in the range of 1-5% RSD. An example of an extracted ion flour, corn meal, brown rice flour, breakfast cereals,
chromatogram for commercial milk Brand F is shown in multivitamin tablets, and milk samples. Accuracy of the
Figure 4. The USDA Nutrient Database for Standard method depends on achieving full equilibration between
Reference, Release 19 (23) lists a value of 1.07 ppm niacin for endogenous niacin and the labeled spike, as well as on having

247

a)I28-3'84

.%

J--il ..
1

Time
1,50 2.00 2.50 3.00 350 4.00 450 500 5.50 6.00 6.50 7.00 7.50 8.00 8.50 9.00 9.50

Figure 4. Sample extracted ion chromatograms obtained for commercial milk sample Brand F spiked with
D4-niacin.
 GoI.DscHMIDT& WOLF: JOURNAL OF AOAC INTERNATIONAL VOL. 90, No. 4, 2007 1089
CL
Q.
0.2
0 ---. .-
02 04 06 08
ppm niacin due to addition
Figure 5. Standard additions curve for milk sample
NFYO4O9F6.
accurately known standard niacin solutions. Accuracy of the
method is supported by the results obtained for the RMS. It is
possible to make niacin determinations by this method in SIR
mode for some samples, but for most food matrixes successful
SIR analysis is likely to require more extensive sample
cleanup or a different chromatographic approach than
described here. The extra selectivity provided in MRM mode,
however, greatly reduces interference problems and should
make the method as presented applicable to a wide variety of
food matrixes. The method presented uses an acid digestion,
but the IDMS approach should also be applicable when
alkaline digestions or extractions are used on food samples.
Precision achieved for the samples ranges from 0.5-5%.
Sensitivity is also very good, with LOD near I ppb.
References
(I) Official Methods ofAnalysis (2000) 17th Ed., AOAC
INTERNATIONAL, Gaithersburg, MD, Methods 944.13,
985.34
(2) Official Methods ofAnalysis (2000) 17th Ed., AOAC
INTERNATIONAL, Gaithersburg, MD, Methods 961.14,
975.41, 981.16, 968.32
(3) Tyler, T.A., & Shrago, R.R. (1980)]. Liq. Chromatogr 3,
269-277
(4) van Niekerk, P.J., Smit, S.C.C., Strydom, E.S.P., &
Armbruster, G. (1984)]. Agric. Food Chem. 32, 304-307
(5) Hirayama, S. (1998)]. AOACInt. 81, 1273-1276
(6) Rose-Sallin, C., Blake, C.J., Genoud, D., & Tagliafcrri, E.G
(2001) Food Chem. 73, 473-480
(7) Lombardi-Boccia, G, Lanzi, S., & Aguzzi, A. (2005)1 Food
Comp. Anal. 18,39-46
(8) LaCroix, D.E., Wolf, W.R., & Vanderslice, J.T. (1999)1.
AOACJn. 82, 128-133
(9) LaCroix, D.E., Wolf, W.R., & Chase, GW., Jr (2002)1.
AOACInt. 85, 654-664
(10) Rizzolo, A., Baldo, C., & Polesello, A. (1991)1 Chrornatogr
553,187-192
(II) Saccani, G, Tanzi, E., Mallozzi, S., & Cavalli, S. (2005)
Food Chem. 92, 373 379
(12) Tyler, TA., & Genzale, J.A. (1990)]. Assoc. 0/f Anal,
Chem. 73, 467-469
(13) Juraja, SM., Treneny, V.C., Millar, R.G, Scheelings, P., &
Buick, D.R. (2003)]. Food Comp. Anal. 16, 93-106
(14) Lahely, S., Bergaentzle, M., & Ilasselmann, C. (1999) Food
Chem. 65, 129 133
(IS) Fassett, J.D., & Paulsen, P.J. (1989) Anal. Chem. 61,
643A-649A
(16) Ball, G.F.M. (1998) Niacin and Ttyptophan, Chapman and
Hall, London, UK, pp 319-359
(17) Friedrich, W. (1988) Niacin: Nicotinic Acid, Nicotinamide,
NAD(P), Walter de Gruyter, Berlin, Germany, pp 473-542
(18) Eitenmiller, R. R., & Landen, W.O., Jr (1999) Niacin, CRC
Press, Boca Raton, FL, pp 339-367
(19) Angyal, G. (1985) Production, Regulation, and Analysis of
Infant Formula: A Topical Conference, AOAC, Arlington,
VA, pp 152-159
(20) Neeter, R., & Nibbering, N.M.M. (1971) Org. Mass
Spectrorn. 5, 735-742
(21) Tanner, J.T., Angyal, G, Smith, J.S., Weaver, C., Bueno, M.,
Wolf, W.R., & lhnat, M. (1988) Frezeniu.s' Z. Anal. Chem.
322,701-703
(22) Chen, P., & Wolf, W.R. (2007) Anal. Bioanal. Chem. 387,
2441-2448
(23) U.S. Department of Agriculture, Agricultural Research
Service (2006) USDA Nutrient Database for Standard
Reference, Release 19, Nutrient Data Laboratory Home Page,
http://www.ars.usda.gov/nutrientdata