Chromatography Forum

Doing a project on a reverse phase HPLC analysis of 5'-mononucleotides. Currently using a single point calibration. What are the benefits of upgrading to a 5-point calibration curve?

Thanks

Um... increased accuracy and precision?

Essentially you are doing a two point calibration at the moment, your single point is your known standard with a two point curve running through 0. At least this way your RSQ is always 1. If your method works as it is, why change it? Although I always think a 1 point curve is a little "iffy", even in a non-GLP environment.

But you could also start to find problems with your method. i.e. what happens if your calibration curve is not linear?

Doing a project on a reverse phase HPLC analysis of 5'-mononucleotides. Currently using a single point calibration. What are the benefits of upgrading to a 5-point calibration curve?

Thanks

The first response pretty much sums it up. The question I have is why go from one ( or two, counting zero ) to five?.

You could just ensure that you bracket your samples with a std above the highest sample concentration and second std below the lowest sample concentration. That makes two or three ( counting origin ) which would provide a significant increase in confidence.

The more standards the more confidence, but also each additional standard offers diminishing gains, and has more potential for discovering problems that are also increasingly trivial..

If it's not broken, don't try to fix it.

Bruce Hamilton

Or if you don´t know its broken you don´t need to fix it?

With one point + zero you have to assume that the line is straight and that it goes through zero on both axes.

With two points you have to assume that the line is straight.

With three points in a row the line could be straight or S-shaped.

With four points in a row you have to start using high order polynomials to fit anything except a straight line.

etc

It is important that the points are evenly spread - it is a standard dirty trick to have a cluster of scattered points in the lower left corner of the graph, and a single point at a much higher concentration in the top right. This gives a high R squared, but effectively has only two points on the line.

Peter

With one point + zero you have to assume that the line is straight and that it goes through zero on both axes.

It's my experience that calibration curves almost never go through 0,0... Standard practice for us is not to force it through 0,0... But what is the most correct thing to do I don't know..

If the 5 point calibration curve is proven to be linear during method validation, the a single point calibration for routine analysis is adequate.

Sorry, ntruong. I can't agree that linearity of a five point line (or any other number of points) validates a single point calibration forced through 0,0.

The calibration:

concentration (in whatever units) = (100 x peak area) + 100

is perfectly linear, but it does not go through 0.0.

in the general case of y = ab + c, the line only goes through the origin if c = 0. In the real world where you always have some background signal, or some small mass of analyte that you cannot detect, c is always either +ve (background) or -ve (undetected analyte)

Peter

Peter, good point, but two things are not clear in my mind.
1) It seems that if the background is constant then it cancels due to peaks being calculated from the baseline.
2) If you do a standard, lets say, every 10 unknown samples (or even use an internal st. in every unknown sample in addition to the external st.) and if this standard´s concentration is nearly invariate you are doing sort of an one point calibration which should be ok for quite a few injections?

If the standard varies then you might have the problem discussed in "Performance qualification detector response intercept", or gunk on the column retains some analyte "permanently", or.....? So if the st area (hight) varies maybe one should do a 5-point again to see what gives?

If the original poster is using UV, I'd be very surprised to see S-shaped curves ( or many other glorious shapes ) appear in most methods.

My experience is that calibration curve problems occur most often at wavelengths where mobile phase components start absorbing strongly, or when samples contain other material. If the samples are consistent, and the method is sensible and robust, calibration curves should be consistent. Batch to batch variations of mobile phase components will have an effect, but are easily controlled, as are changes in column behaviour.

I'd also like a little clarification on the comment by ntruong about single point versus 5 point. If you produce a calibration curve of 2.0, 3.0, 4.0, 5.0 units during method validation, and you plan to use the response of 4.0 standard +- some error as your single point. You could still use that calibration curve, and there should be no need to force through zero. Am I missing something implied.

Do people really have so much instrument time available that they spend time running 5 point calibration curves even when they are linear?. It's about time somebody introduced risk assessment and value of instrument time into the equation.

Bruce Hamilton

Now it's getting interesting !!

When I said background, I was thinking about background levels of the analyte, that DO appear as a peak on the chromatogram, rather than background on the chromatogram which just gives a higher baseline. Background analytes are often a problem with ubiquitous contaminants such as chlorinated pesticides or PCBs - it is impossible to get a blank matrix, and so even the "blank" has an analyte peak on it.

One case in which there is "deliberately" a background is when the detector response is calibrated by spiking standard at multiple levels into the sample - this is sometimes the only way to account for odd matrix effects. The concentration in the sample is then calculated as the y intercept divided by the (straight line) slope.

I think that running single point standards amongst the samples is to check for bias rather than for a change in slope.

I would also be surprised to see an S-shaped curve in chromatograhy as long as the calibration range was sensible, but a three point line cannot rule it out.

Once linearity has been established, multi-point calibrations are used to verify performance over the range. With three points or fewer a deviation from linearity could be due to any one of the points, with four or more you can see which point is off the line. The need for more points comes from the probability of getting a straight line (or high R squared) by chance with few points (with 2 points it is 100 % !), even with 4 points you can get an R squared of 0.99 with one point well off the line.

The more rigourous guidelines now want 7-point calibrations, throw in blanks and check standards and the actual samples are often in a minority !

Peter

If you have a validated method and experience, single standard should be OK.

Given the above, the added benefit of a multipoint curve is that it allows you to "see" some of the standard prep. errors. The proximity of each point to the trend line and the value of the r² can be indicative of the quality of the standard prep. process and analyst. This aspect is lost when you are limited to 1 standard prep. & 1 sample prep.

Peter, I am not very sure about what you are saying. You seem to be talking about carryover, and or overlap of dirt (dirt being any peak which is not due to analytes). If these interferences are not constant (they seldom are in my experience) than it´s impossible to determine them in order to subtract them from the results of unknown sample determinations. You have to do a better or different chrom., or something alltogether different. So what is the point of doing this multiple level spiking?

Hi HW Mueller (forgive the formality).

Once again I am not making myself clear.

I am not talking about carryover or interfering contaminants (dirt), which I agree must be eliminated by sample clean up, better separation, or detector selectivity.

Having to spike a sample with multiple levels of analyte most often arises in headspace analysis, where the partition ratio of the analyte(s) into the headspace can be strongly dependent on the composition of the matrix. Because only the headspace is analysed, the peak area for a given analyte is also stongly dependent on the composition of the matrix.

Consequently, to calibrate the analysis it is necessary to make up standards in a matrix that is the same as that of the samples. Ideally this standard matrix would not contain any of the analyte. Sometimes such analyte-free matrix does not exist, or if it does exist it is difficult and/or expensive to produce.

In these cases the most effective approach is to use the sample itself as the standard matrix. The sample is run as is to find the peak area due to the analyte already present. The sample is then spiked with analyte at enough different levels to establish the shape of the calibration curve (I would say three levels as a minimum - giving four points). The best fit line cuts the y-axis at or very close to the unspiked peak area. Depending on your taste you can take either the intercept or the area to calculate the concentration in the original sample.

This process is most easily carried out using separate portions of the sample for the different spiking levels, but if needs must (and with some extra allowances for analyte lost during sampling) the whole process can be run on one portion of sample.

In principal this approach could also be used for more conventional liquid extracts, but for such samples it is nearly always more effective to use internal standards, or determine recovery and account for it in the calculations.

The major disadvantage of spiking into samples is that each sample has to be run four or five times. I only do it when there is no alternative.

Regards Peter

If somebody presented me with a routine HPLC-UV method for a seven point calibration curve for a single analyte, I'd probably tell them to go elsewhere, as the price to them would be "talk to your bank manager" territory.

If you want a sensible sevn point calibration, the number of replicates probably should also increase. Besides, accepting seven point methods legitimises such nonsense. Customers may legitimately believe the process is being deliberately padded out.

Setting up for single research samples can already bite hard ( two blanks, two standards, blank, sample, blank, two standards, blank ), ten injections fior one data point, and injections may normally be duplicated.

Somebody should have looked at the method and told the author it was unacceptable, rather than approve it. There are instrumental methods ( eg AAS ) and even HPLC detectors ( ELS, MS) where more standards may be justified for difficult samples, but HPLC-UV?.

Bruce Hamilton