Loading Capacity of HPLC Columns

[Bob Albrecht]

I need to know the loading capacity of various columns, in particular I would like to understand the sample loading capacity of Waters UPLC columns. We want to use them for micro-scale preparative separations for natural product elucidation.

Can anyone tell me how much sample can be loaded on a UPLC column, and if there is a calculation for capacity vs. particle size, surface area, diameter, etc.

Thanks,

-BOB-

[SIELC_Tech]

Here is study on loading capacities for Atlantis T3, Zorbax SB-AQ and Obelisc R columns for dopamine -basic hydrophilic compound (although this is not UPLC but it is scalable to any particle size):

http://www.sielc.com/Products_Obelisc.html

Here is Bristol Myers poster for comparison of reverse phase, mixed-mode and HILIC:

http://www.sielc.com/pdf/SIELC_PrepOptimization.pdf

and another piece of info:

http://www.sielc.com/Technology_Prepara ... raphy.html

Mixed-mode columns usually provide several folds increase in loading capacity for ionizable analytes.
Also there is publication by David V. McCalley, in Journal of Chromatography A, Volume 1138, Issues 1-2, 5 January 2007, Pages 65-72.

David McCalley published several articles on loading capacities of reverse phase, mixed-mode and HILIC columns.

P.S. I represent manufacturer of mixed-mode columns and my opinion might be considered bias by some participants.

[JA]

reload..reload..reload..reload..reload..reload..reload..reload..
reload..reload..reload..reload..reload..reload..reload..reload..

I'll put my hand in and have a go at this one while we wait for the resident expert on Waters packing materials and prep LC

I haven't come across a standardised test for column capacity and thus I don't think that the information you have requested is actually out there. Sure, one can acquire data on the physical properties of various stationary phase packings but it will not include reference to their sample capacity. Users often assume that this capacity increases in proportion to variables such as the particle surface area, ligand chain length or carbon load which is not necessarily the case when comparing different packings, e.g. a C18 from two manufacturers.

You can calculate the sample capacity of a given column assuming you have already determined the capacity of one which encorporates the same column chemistry. This is one of our fundamental approaches in preparative chromatography: determine the loading on an analytical column and scale up in proportion to the change in column volume. If we decrease the particle size, I'd expect the capacity should go up in proportion to the specific surface area assuming that the ligand bonding density remains unchanged.

One way to measure the column capacity would be to spike a known concentration of a test probe into the mobile phase and measure the breakthrough time. I'm not sure if this will yield publication-quality data but it should let you make informed comparisons.

In summary, determining the sample capacity (loadability) is very much in your own hands. It is entirely application specific. Preparative chromatography is not governed by the appearance of pretty narrow and symmetrical peaks and is usually performed in overload conditions. The only criteria for a successful project is one where the product is recovered with satisfactory yield and purity.

[JA]

The dopamine is a clever example because it is highly polar and permanently ionised; it isn't simply 'basic'. Both of these properties wouldn't necessarily apply to a large number of pharmaceuticals and the comparison is even weaker to the kind of molecular features I would ascribe to 'natural products'.

Was the project work leading to the BMS poster sponsored by SILEC? The sample identities are not disclosed, only that they
- are "highly basic ionic compounds" (permanently charged?)
- "do not elute under normal phase conditions and show little retention by RP" (you mean highly polar)

Do we have a comparison of loading capacities for products which are not model-compounds on the SILEC phases?

[SIELC_Tech]

JA,

Poster was not sponsored by SIELC (we are not big enough to sponsor big pharma)....we just know people from BMS from our days at Pfizer.

Higher loading capacity for mixed-mode is only true for ionized compounds. For neutral compounds will have same capacity.
Also if you go to much higher pH (10-12) you will have most basic compounds non-ionized and retained by reverse phase mechanism. I don't know what effect residual silanols will have on this. If natural compounds are non-ionic you can go with any reverse phase column, but some of them might be charged compounds (anthocyanins, amino acids, etc.).....and then you need to decide which way to go

[Bryan Evans]

There was an interesting scientific poster done at ASMS 2008
that relates to this topic:

"Evaluation of Column Retentivity with large injection volume
for High Sensitive and High-Throuput LC-MS/MS Quantitative Analysis."
J. Watanabe, et. al. Takara Bio - Japan

If anyone would like a copy of the poster - send me an email and
I can forward the link on to you.

[Uwe Neue]

The load that you can put on a column is not a question of particle size. With other words, the adsorption isotherms are the same for a 1.7 micron particle and a 10 micron particle.

To get a feel for the lodability of different packings, calculate the ratio of the specific surface area to the specific pore volume. This is better than using just the specific surface area, which is misleading.

Most importantly, the loadability for an analyte depends on its state. If you have ionizable analytes, you get a 10- to 50-times higher loadability if you get them into the unionized form, either at acidic pH for acids or at basic pH for bases.

[emsnyder]

Uwe,

Why would you need to use the ratio of surface area to pore volume? So long as the molecule is small enough to access the pore (100A is more than enough for small molecules), the entire surface area is accessible. I have done a fair bit of prep chromatography and loadability scales quite well with surface area. Dividing by pore volume will give you a nonsense result.

I know you work for Waters, but I have seen numerous real data (not marketing) for high pH loading of basic compounds (above pKa) on Gemini versus XBridge prep columns. The Gemini have approx 2x loading of the XBridge. And the surface area is approx 2x (375 m2/g Gemini, 185 m2/g XBridge).

I agree with you 100% about the ionized states. If you have acidic compounds, use a high surface area silica column - Waters Sunfire, Phenomenex Luna, Supelco Ascentis, etc. at a low pH.

If you have basic compounds, I would go with a Phenomenex Gemini-NX at a high pH. I am not aware of any other high pH stable columns with a large surface area.

BTW, I do not work for any column manufacturer. I work at Battelle in OH.

[Uwe Neue]

The reason that you should not look at the surface area alone, but at the surface area per pore volume is simple: if you have a packing with a low pore volume, you can pack more mass into a column, and the amount of surface increases with the mass of packing in a column. This is why I argued to look at the specific surface area per specific pore volume instead of looking at specific surface area per gram. If you do that, you will realize that packings with 200 m^2/g and 0.5 mL/g or 400 m^2/g and 1 mL/g have about the same ratio of surface area to pore volume, and the same pore size.

In order to be precise, you need to do a slightly more complicated calculation. To estimate the amount of surface in a column, you need to include the silica, since the silica does occupy some volume in the column. Assuming for simplicity that the density of silica is 2.0, the ratio of surface area to particle volume is 200/(0.5+1/2) = 200 in the first case and 400/(1 + 1/2)=267 in the second case. As you can see, this is not a 2:1 ratio (as a simple look at the surface area would have indicated), but a 1.33:1 ratio. This is what I wanted to point out. Of course, you are also correct that I simplified somewhat, when I suggested to look only at the ratio of surface area to pore volume, but you also see that the latter is closer to the truth as just looking at specific surface area.

As you can see, this is not nonsense, but clear logic. I did NOT argue about the materials from different vendors.

[emsnyder]

Uwe,

I understand were you going, and I was too strong in saying "nonsense", but I stand by what I say in that pore volume is inconsequential to the packed density (mass of silica per unit volume).

In your example, the lower suface area particle has 75% of the surface area of the higher (200/267=75%). This is just as inaccurate as not including pore volume (200/400=50%), or simply dividing by pore volume as you suggested (400/400=100%).

But your example is even more inaccurate because it is a theoretical model and does not include several real factors. You're assuming that the volume of silica in the column is either pore volume or solid silica volume. What about the interstitial volume? This is not insignificant and must be included, particularly with larger particles often used in prep LC. Also, a column is not a perfect hexagonal close-packing of perfect spheres (unless we want to be physicists).

If what you are saying is correct, a column column packed with material with a pore volume of 0.5 mL/g would have 50% of the void volume of a column packed with material with a pore volume of 1 mL/g. This is clearly not true as the void volume (or time) does not differ that significantly with columns packed with different materials.

I have never unpacked different columns and weighted the amount of material in them, but it would be interesting to do. I can't find much data on this either. The Phenomenex catalog has the amount of some different materials required to pack a DAC column. Here is what they state (the format is: material, pore volume in mL/g, packing density in g/cm3).

Luna 10um C18 100A, 1.0, 0.566
Kromasil 10um C18 100A, 0.9, 0.666
Lichrospher 12um RP-18 100A, 1.25, 0.647
YMC 10um ODS-A 300A, 0.9, 0.503
Vydac 10um C4 300A, 0.6, 0.580

Do you see a pattern? I don't. If I look at the Luna and Kromasil, it makes sense, Kromasil has a smaller pore volume and a higher packing density. But look at Lichrospher. It has a larger pore volume than both and it is a larger particle, but its packing density is almost the same as Kromasil. Same with the 300A stuff, similar packing densities even though the Vydac has a significantly lower pore volume.

So I stand by my assertion that with real materials, the pore volume does not make a difference. If you have the spare time and columns laying around, unpack and weigh the materials and let us know the results.

[Uwe Neue]

The reason that you do not see a pattern in the data that you show is because you are comparing a hodgepodge of bonded phases from C4 to C18 on different pore sizes, different bonding densities etc. Plus, data on how much packing is required to pack a DAC column is furthermore complicated by particle strength, particle shape, and is a very fluffy number to start. If you want to dispute this, I can support all these statements.

But let us go back to the original argument. The reason that my examples have been chosen correctly is the fact that the correct average pore size of a packing is the ratio of the specific pore volume (which can be determined without too many problems) to the specific surface area (which is easy to measure). The reason behind this rule (which goes back to Klaus Unger) is essentially the similarity of pore structures. You can readily imagine this yourself by looking at this ratio for similar structures: spheres, channels etc. While the value for this ratio is different for a sphere and a channel, it is characteristic for self-similar structures, and one can argue that silicas are self-similar structures.

So if you analyze my examples, you will see that they have been chosen to have the same pore size according to the statement above: 400 m^2/g at a pore volume of 1 mL/g versus 200 m^2/g at a pore volume of 0.5 mL/g.

The rest is straight algebra.

The story about the void volume is more complicated than what you have stated. The void volume consists of two components: the interstitial or inter-particle volume, and the pore volume. Most columns have an interstitial volume around 40% of the column volume. The remaining 60% are occupied by the particles. And the particle volume consists of the pore volume and the true backbone, which carries the surface. The discussion above applies to the particles alone. Therefore it is not correct to conclude that a column with a pore volume of 0.5 mL/g has 50% of the void volume of a column packed with a material of 1 mL/g.

[emsnyder]

I think we are going to have to agree to disagree on this. To quote you:

"The reason that you do not see a pattern in the data that you show is because you are comparing a hodgepodge of bonded phases from C4 to C18 on different pore sizes, different bonding densities etc. Plus, data on how much packing is required to pack a DAC column is furthermore complicated by particle strength, particle shape, and is a very fluffy number to start."

Exactly my point. There are many factors when trying to "calculate" packing density. I can think of a few others you didn't mention above. My point is that the pore volume is just another one of these factors - some of the other factors may cancel out each other, some may not.

I am not sure what you are arguing about the pore size. My assertion is that the loadability of a column is a function of its surface area and packing density. And that your estimate of packing density by using pore volume does not work - e.g. packing density is an empirical measurement based on many factors about the material and manufacturer, and pore volume is just one of these many factors.

So I stand by my assertion that the best "guestimate" of loadability is by comparing the surface areas, and not including the pore volume.

I am sure you are busy, but if you or a colleague at Waters has a few minutes, take a Spherisorb (0.5 PV) and an Atlantis (1.0 PV) and unpack them and weigh the material. If you have even more time, try to do a few more (Sunfire, Symmetry, Delta-Pak, X-Bridge, etc.) to attempt to get a statistical representation of different pore volumes and surface areas, while holding constant the particle size, ligand, and manufacturer.

[Uwe Neue]

I am sorry that I have to disagree with you, but I can not leave an argument where I know that 1 + 1 = 2, while you declare that the result must be 3.

The phase ratio P of a packed bed is:

P = Asp/ (Vsp +Ei/(1-Ei)*(Vsp+1/rho))

Asp is the specific surface area of the packing. Vsp is its specific pore volume. Ei is the interstitial fraction, typically 0.4, and rho is the density of the particle backbone, which is 2.2 g/mL for silica.

You can try to put in the values for the two packings mentioned above, which have the same pore size (400 m^2/g and 1 mL/g versus 200 m^2/g and 0.5 mL/g). You will get 203 m^2/mL of packed bed in the first case and 176 m^2/mL of packed bed in the second case.

As you can see, the difference is really small.

If there are loadability differences between different packings in real life, they are related to issues other than the specific surface area. It is important to keep this straight.

[emsnyder]

OK, I surrender. I will agree with you that this is true in theory.

But the relationship between pore diameter, pore volume, and surface area is more complicated for real materials. You would expect particles with the same pore diameter and pore volume to have the same surface area, right?

Both Atlantis C18 and Luna C18 have a pore diameter of 100A and a pore volume of 1.0 mL/g. The Atlantis has a surface area of 330 m2/g, but the Luna has 20% more surface area at 400 m2/g.


It reminds me of an old joke:

A physicist, a nutritionist and geneticist were arguing about how to get the best racehorse.

‘You have to breed for the right characteristics,’ says the geneticist.
‘No, you should get the diet just right,’ says the nutrionist.
‘Well,’ says the physicist, ’let’s imagine the horse is a sphere on a frictionless track ...’

[Uwe Neue]

Well, part of the problem with the data is the measurement. One can determine the specific surface area without any problems. One can determine the pore volume with little problem, but one needs to define, where the pore volume ends and where the interstitial volume starts. This may be different in different companies. The worst thing is the pore diameter. It is completely meaningless, since all values are nominal values to start, different techniques are used (BET or Hg porosimetry), and there is no definition of the center of the distribution, nor if the center should be used or something else.

Based on the specific surface area, which is acurate, and assuming that the pore volume is defined the same way at Phenomenex and Waters (which may not be), my conclusion is that the pore size of the 100 A Luna packing is smaller than that of the 100 A Atlantis packing.