Try destroying any citrate in your oranges before derivatization, then use it as your blank.

I would go with Bruce's suggestion. You may be able to get citrates out by macerating the sample and hydrolyzing/extracting with NaOH and filtering. Then wash the orange with water or dilute HCl to get the residual hydroxide out; then wash and filter again. You could then use that as a blank, You could freeze and store the extracted pulp for later experiments. Obviously with a technique like QuEChERS you run the risk of removing some of the other potential matrix interferences as well...

Alternatively "standard additions" "spiking" can be done for an analyte with difficult, or undefined matrices. The method is not perfect, but can be very useful. The procedure is to do a baseline "blank" on your sample. Typically several duplicate samples are run to determine a baseline (As the analyte is actually in your sample, this is not a real "blank"). Then spike several duplicates with a known amount of your analyte of interest to determine whether you can "see" it over the baseline "blank" value; and if so, determine the precision. If still OK, spike at different addition levels and see if you can get a calibration curve. Then, depending on the analytical technique, you can extrapolate the curve to get a baseline zero - The curve might be "x", ("x"+y), ("x+2y), ("x"+4y) etc. - The difference between the samples with additional standard spikes (y) and actual zero, and the calculated one from your curve, will be the actual amount "x" in your sample . You would normally have to do this with representative samples from each batch, so it can get very tedious. If you are lucky, you will have several matrix samples that you can use containing known amounts of analyte.

With techniques like QuEChERS by, say GC-MS or HPLC-MS, there is a trick that you can do with "real" samples. This involves taking a known small amount of clean solvent into a syringe barrel - Lets say 1µL of pure acetonitrile. Pull the solvent up into the barrel and now load the syringe with 2µL of sample extract and draw that into the barrel. You should have two samples of liquid (For a GC they are normally separated by a small air gap). Inject this into the instrument. This is your "blank". Now repeat the experiment with the same extract, but this time replace the clean solvent with acetonitrile containing a known amount of citrate. Repeat with different concentrations of citrate.

Some modern syringe autosamplers can automate standard addition - They allow "sandwich" injections which are typically pure solvent, a segment containing your standards, and then a blank matrix sample (or sample extract). The method can also be used for adding small amounts of expensive Internal Standards to the middle segment with a high degree of precision; but there is no reason why you can't use this for a spike and put an internal standard into the "pure" solvent. Some systems can even be programmed to produce different concentrations by calculating different volumes of the standards solution, and clean solvent (If you use an IS as well as the spike, and the IS is in the first segment, its concentration will vary as well -  So your calibration calculation must be adjusted - To avoid this, manually add the IS to the extract).

As Tim says, modifying representative samples to remove the endogenous metabolite can have limitations.

The method of standard additions does not solve the problem either because the question, rephrased, concerns primarily the specificity, and not matrix effects on the response.  We need first to establish whether the analyical response without standard addition is due entirely to the analyte citrate. Standard additions deals only with the question of whether the analytical response or working range is affected by the matrix. This situation commonly arises with LC-MS-MS methods of impeccable specificity, whenever a stable-isotopically labelled internal standard is not available.

The "perfect" solution is to dispose of a reference method for which we can be sure of the the specificity; for a major metabolite like citrate, single-column GC-MS should easily meet this requirement. Given the separating power GC columns, a non specific detector like FID may be judged sufficient. Possibly, suitable reference data for your analytical situation may already exist in the literature.

Many laboratories don't have access to expensive facilities. The best you can do may be to compare results using more than one analytical method, for example two kinds of liquid chromatography (which for the present problem can be found in manufacturers' catalogues). The decision on how far to take such development work is often dependent on economic factors, and it can sometimes be legitimate to give results as "citrate by such-and-such a method" in order to indicate to the client that there is a small risk of uncertainty due to lack of specificity.

Christopher makes a good point. As Chen's "Topics" are LC-MS and HPLC, I had made assumptions (perhaps incorrectly!) that the techniques used would have a relatively high specificity - Unlike, say, GC-FID or HPLC/single wavelength detection - Hence my inclusion of standard addition which, if the technique has a specific enough spectral response for the analyte, may be used to estimate endogenous analytes as well as for evaluating detector responses/working ranges (as normally used for exogenous compounds).

With these methods, we would be looking at spectral purity within a retention time window - A library search match with an MS in scanning mode; or more likely, a screening SIM method with a quantification ion and several qualifiers; or a DAD for HPLC. In these cases standard addition would show up "unexpected" quantification and "purity" differences from interfering contaminant compounds, as well as being useful in determining matrix effects and working ranges. I agree that we would expect that the additional specificity of MS/MS methods would be better at showing these differences.

As Bruce says, derivitization can help specificity (as well as its normal use of improving chromatography). If you really want to be certain of distinguishing your analyte when looking at possible interfering compounds, there is no reason why you should not look at running separate experiments with different derivatives (like methylation vs. trimethylsilylation) and compare the concentrations obtained. Another approach, would be to repeat the experiments with a different polarity column (GC or HPLC) or a different mobile phase (HPLC).

Thanks to Tim and Christopher. Acoording to these suggestions, the design of my validation experiment should not merely following arrangements of others, instead, actual issues, such as specificity and matrix effects should be concerned respectively. The strategy of multiple analytical technologies offered by some third-part lab is perfect to me.