Handling reactives in the solubility database

The discovery that aromatic aldehydes bearing electron withdrawing groups react with alcohols has forced us to create a new category in the ONS Challenge SolubilitySum spreadsheet: solute reacts with solvent. There is now a column on the far right to flag these.

Since there is no meaningful number for these mixtures we are assigning the value of -0.1M in the solubility column of the SolSum sheet. That has the advantage that it will not break any web services currently querying the sheet. We can also put an explanation in the notes column for a reason, such as "forms hemiacetal".

One of the benefits of this approach is that all solutes reacting with solvents can be queried using Rajarshi's web query setting the solubility range from -1 to -0.01 M.

Solubility surfaces in 3D

After a little bump in the road with attempting to measure the solubility of 4-nitrobenzaldehyde in methanol, Marshall Moritz extended the study (ONSC-EXP111) to chloroform and acetonitrile, where no hemiacetal can form. The sequential precipitation method seems to work quite well with a new thermostated bath that lets us go down to almost -30C.

The values in pure acetonitrile and chloroform were consistent with those obtained from Maccarone, E.; Perrini G. Gazetta Chimica Italiana. 1982, 112, p. 447. (private access). But Marshall went further and measured 9 points with mixed solvents and different concentrations. The result is this saddle shaped 3D surface plot.


I think the potential for extending the ONS Challenge to cover the full multi-dimensional space of mixtures of about 10 common solvents and temperature is very exciting. It would be foolish to think that we can map in extreme detail such large surfaces. However, I'll bet we can come up with some useful estimates of what the surface looks like in many areas of the space.

The long rage plan for this approach would involve coming up with at least an empirical model based on molecular descriptors of the solvents and solutes. This would help predict the space for completely new combinations.

Being able to predict the solubility of all starting materials and products would enable organic chemists to rationally select solvent systems and temperatures for their reactions maximizing product yield from simple precipitation. This is something that we are investigating for the Ugi reaction and I'll report on this as the data come in.

NMR integration progress for solubility measurements

In a previous post I reported about issues with using H NMR integrations for quantitative assessment of solute concentration to determine solubility. Using default parameters, a study by David Bulger revealed that integration of the methyl group on 2-phenylbutanoic acid lead to only 8.4% average error while the methine H gave an average of 26% under-integration.(ONSC-EXP102)

We have tried to select methyl or methylene groups whenever possible but obviously this won't work for many compounds, especially our series of aromatic aldehydes and carboxylic acids.

Khalid Mirza has just done a study using longer relaxation times and the results appear very promising (ONSC-EXP103). For 4-pyrenebutanoic acid using default parameters the integration of the aromatic vs. methylene hydrogens was off by 9-18% while extending the relaxation time (d1=50s) reduced the error to less than 1.2%.

Furthermore, Marshall Moritz used the new settings in a measurement of a known concentration of 4-nitrobenzaldehyde in acetonitrile and came within 3.2% using an aromatic hydrogen (ONSC-EXP111). Use of the aldehyde hydrogen was not as good, although much better than what we normally see with the default settings.

So I am cautiously optimistic that we are likely to get better data going forward.

Complications with solubility measurements of aldehydes in alcohols

Last week I mentioned that we had started measuring temperature solubility curves. Since 4-nitrobenzaldehyde in methanol also got flagged by our Outlier Bot we picked it as the next one to process. This one has bothered me for some time because our values have systematically come out to about half of those published by Maccarone, E.; Perrini G. Gazetta Chimica Italiana. 1982, 112, p. 447. (private access) In this paper temperature curves for the solubilities were also provided and so I thought this would be a great opportunity to validate our sequential precipitation method.

This experiment (ONSC-EXP110) turned out to be very puzzling. On the day that we tried to make solutions we could not get as much material to go in as expected from the paper, even by sonicating at 40C. The next day some of these samples went into solution and then we could not get precipitation to occur for some samples even at 0C.

However, an examination of the NMRs of the samples with lowest concentration revealed the problem. It turns out that they were about 50% converted to the methanol hemiacetal. The NMR clearly shows the coupling between the benzylic and hydroxyl protons at 5.6 and 6.8 ppm, each integrating for 1H each compared with 2H for each of the new peaks in the aromatic region.



It turns out that the hemiacetal was present in all our previous measurements for 4-nitrobenzaldehyde in methanol but we didn't spot it. Furthermore we found significant hemiacetal formation for 2-chloro-5-nitrobenzaldehyde, 2,6-dichlorobenzadehyde and 4-chlorobenzaldehyde but only trace on none for others like veratraldehyde (ONSC-EXP033 has several of these NMRs). So it appears that if the aromatic ring bears sufficiently electron withdrawing groups hemiacetal formation is facilitated without the presence of a catalyst. This makes sense from a mechanistic standpoint.

Since the hemiacetal forms so easily in these cases I wonder why there was no mention made of it in the Maccarone paper. They assayed concentration using GLC - perhaps the hemiacetal decomposed quickly under those conditions? (PNBA is para-nitrobenzadlehyde):

From a practical standpoint is it even possible to measure the solubility of 4-nitrobenzaldehyde in alcohols? If you try to make up a solution for a reaction the concentration will be much less than calculated. With an excess of alcoholic solvent it seems unlikely that a reaction consuming the aldehyde will shift the equilibrium enough to make the formation of the hemiacetal inconsequential.

I think that for the Ugi reaction the implication of this is that alcoholic solvents should not be used for aldehydes bearing electron withdrawing groups.

On the Advisory Board of Chemical and Engineering News

I have accepted an invitation to join the Advisory Board of Chemical and Engineering News. The three year appointment involves yearly meetings in Washington, D.C. where important trends in chemistry and the focus of the magazine are discussed.

Many thanks to Carmen Drahl for putting my name forward!

BrightTALK on ONS and Drug Discovery

I will be presenting at the BrightTALK seminar series on July 23, 2009 at noon ET. The virtual presentation will be on "Open Notebook Science for Collaborative Drug Discovery".

Temperature solubility curves using sequential precipitation

We were going to wait for a while to tackle temperature related solubility because it wasn't clear that our NMR technique was applicable. The main problem is that separating out the supernatant at a given temperature can be tricky for both filtration and centrifugation with our available equipment.

However, while discussing the issue with Marshall last week it occurred to me that we might be able to get the data by sequential crystallization. The idea is to prepare several solutions at known concentrations at a high temperature then let the bath cool down and note the temperature when precipitates are first observed. We can then intrapolate to report the solubility at room temperature if necessary.

Khalid and I made an attempt on Friday for pyrene in acetonitrile (EXP109) and obtained a value of 0.16 M at 22C, about twice the value found by Marshall in EXP108 using NMR. Part of the discrepancy is probably related to longer relaxation times of aromatic protons but we have to keep in mind that NMR is not particularly precise at such low concentrations. Hopefully we will soon figure out how to run our NMR experiments to allow full relaxation of all protons and address this issue more conclusively.

Crowdsourcing solubility requests from bots and people

Now that we have a reasonably routine way of measuring non-aqueous solubility and students trained to do the measurements, we can think about adding some structure to the workflow. By using an open Google Spreadsheet to list the next solute and solvent to measure we can easily crowdsource the requests. This is the dosol sheet and students are instructed to check it before planning their experiments for the day.

Requested experiments come from the following sources:

1) OUTLIER BOT: Andrew Lang has created a neat service that reports measurements exceeding a provided standard deviation to mean ratio (plug into the URL) and a Grubbs outlier threshold. Currently, the "bot" does not automatically write to the dosol sheet but eventually I see that it would make sense to complete that integration. On a daily basis I run it and process the flagged entries in the following ways:
a) If I investigate the corresponding lab notebook pages and find an error in the calculations I simply fix it.
b) If I find that one or more of the measurements are obviously in error I mark the entries as DONOTUSE in the SolSum sheet. This is often a low value that was obtained early on in the project before we appreciated how much mixing is required for some solutes.
c) If I cannot determine why there is a discrepancy I place a request on the dosol sheet. (note: solubilities that are very low (<0.1 M) may have relatively high standard deviation to mean ratios because the techniques we use are not precise at those solubilities and remeasuring won't help)

2) Internal Requests: If someone from within our group is planning a Ugi reaction and there is a missing solubility measurement for one of the reagents or Ugi product, this is a convenient way to request the missing information to make a solvent choice for the reaction.

3) External Requests: When people think about crowdsourcing, usually the first thing that comes to mind is how others can help solve their problem. But in addition to asking what the science world can do for you why not ask what you can do for the science community? We have done so last week and received 3 requests (via a form created by Andy). One for the solubility of iodine in 1,1,2-trichloroethane, one for the solubility of nitric acid in water and another for pyrene in acetonitrile.

For the first request, iodine has no hydrogens so we can't use our standard NMR technique to do it so I set the priority of that request lower. For nitric acid - it is well known that it is miscible in water so we won't be doing that experiment. We also want to focus on non-aqueous solubilities. However the third request was a good fit and Marshall Moritz from our lab processed it right away - he found a value on 0.08M for pyrene in acetonitrile at room temperature (EXP108).

Hadas Joseph from the Agricultural Volcani Center in Israel made the third request. He is a masters student investigating the degradation of pyrene in soil and needs the measurements for extraction experiments. The contacts we make via this route might turn into more involved collaborations or just end up being interesting people to interact with. Either way it benefits science.

4) Implicit requests by keyword searches: We know from our Sitemeter and Google Analytics the keywords that people use to find our site. Most of the time they find the solvent and solute combination that they are looking for but occasionally a combination arises that we have not yet tried. In fact we set up this page with all of the compounds and solvents in my lab to catch new combinations that we could run. This is how the request for phthalic acid in chloroform found its way on the dosol sheet. If someone or something searched for it there is a need for it to exist.

Ultimately, whether entered by human or machine agency, the requests on the dosol list are simply that - requests. It is still left to me to assign the priority in a logical way, allocating resources in proportion to their importance in the portfolio of projects that we are pursuing. One of those projects involves listening to what the global chemistry community is asking for and betting that it will be useful to answer.

Recent insights about solubility measurements

Over the past several months we have learned a lot more about measuring non-aqueous solubilities. (see experiment list) Here is a summary:

1) Sonication: It can be really difficult to reach saturation for some compounds. Since we started the ONS Challenge project in the fall of last year students have used all kinds of methods to mix the solvent and solute. For a while we required that solid remain during 10 minutes of vortexing. However, this proved problematic for doing lots of samples in parallel and required the researcher to stand there holding the vials on the vortexer.

So to be able to leave samples mixing unattended we started using a sonicator. This certainly seems to be a major improvement over vortexing. But, as an unintended benefit, sonication also caused the temperature to rise. Upon cooling to room temperature, additional solid is often observed to come out of solution - the only sure guarantee that saturation has been reached. Our current protocol is to continue to add solute until 30 minutes of sonication leaves solid in the vials. Temperature of the bath is also taken before and after to ensure reproducibility.

2) Screw Cap Vials: One of the problems with sonication and increasing temperature is that standard one dram vials with caps that snap in often pop out. Using half-dram vials with screw caps completely solved that problem.

3) Filtration, centrifugation and decantation: Coming up with a common protocol to separate the supernatant after saturation is reached is not as straightforward as it might seem. The problem is that saturated solutions can behave quite differently. The easiest behavior to deal with is when the solute just settles on the bottom and simple decantation can be used. This typically happens for low solubility cases where the solute is much more dense than the solvent. In other situations a suspension forms that does not settle quickly. Centrifugation for a few minutes often resolves that issue and then the supernatant can be decanted. However, there are more difficult suspensions that do not settled after centrifugation. The worst of these actually form a gel. We have found that filtering through as Pasteur pipette packed with a small piece of cotton usually does the trick. The gel or suspension is transferred into the pipette and a bulb is carefully squeezed to try to get at least a drop of supernatant out. Even getting a single drop can be challenging sometimes so it is fortunate that our NMR method does not require accurate volume measurements.

4) Evaporation of supernatant: Sometimes we observe higher solubilities than the known saturation values. This was puzzling because looking at the NMRs it was clear that the ratios of the solute to solvent were real. We traced back this problem to the way the supernatant is handled. When the drop is introduced at the top of the NMR tube it rolls down and a significant amount of solvent can evaporate. When the deuterated solvent is then added the ratio of solute to solvent will appear to be much higher than it was in the saturated solution. This is especially problematic for volatile solvents like THF. A similar effect can be caused by waiting too long to take the NMR as solvent does evaporate slowly even with a cap on. The solution to this problem is to transfer the drop of saturated solution directily into a vial containing the deuterated solvent and taking the NMR immediately.

5) Deuterated solvent considerations: We have also had problems with solutes precipitating out when diluted into the deuterated solvent. Polyphenols in DMSO or THF tend to do this when diluted in CDCl3. In this case diluting in DMSO-d6 resolves the issue. If a solvent other than DMSO is used non-deuterated DMSO can be added to DMSO-d6 to conserve deuterated solvent. Its only purpose is to enable locking during NMR acquisition.

6) Additivities of volumes and density predictions: There are two assumptions made in the way we measure solubility - that volumes are additive and that the predicted densities (via ChemSpider) of solid solutes are accurate. If one or both of these assumptions is a dominant contributor of error, we would expect the maximum deviation to occur at higher concentration. To investigate this issue (EXP102) David Bulger made up known concentrations of 2-phenylbutanoic acid and measured the SAMS solubility using the integration of the terminal methyl and the methine proton. Then the error was plotted against concentration.

It turns out that the error does not increase with increasing concentration suggesting that - at least in this system - those two assumptions hold up to 3M.



However there was a marked difference in the error of the methyl (8.4% average) and the methine H (26% average). Clearly the methine proton is systematically under-integrating, a feature of a longer relaxation time. The solution to this in general would be allow sufficiently long relaxation times during NMR acquisitions and we are looking into this.

Still, lets put these results into context. Is a 26% error that bad? Well it depends on the application. The purpose of this project is to give chemists guidelines in choosing solvents for carrying out their reactions. Whether the solubility is 1M or 1.26M is not going to be of much concern to plan a reaction at 0.5 M and rules out reactions at 1.5M.

Also note that very small temperature changes can have very large effects on solubility. For example if we look at the solubility of 4-nitrobenzaldehyde in various solvents between 26C and 19C we find differences of 45% in acetonitrile, 34% in benzene, 59% in carbon tetrachloride, 42% in methylene chloride and 36% in toluene. This temperature range is within the variation of what could be considered "room temperature" in a lab (at least the ones I have worked in).

Sometimes temperatures are reported in solubility studies but often they are not. For example, the solubility of aspirin is reported as 1:5 in ethanol in this book. No exact temperature is given and this is not a very precise measurement. But there is enough information there to plan a reaction or to gain an appreciation for roughly how much solvent to use for a recrystallization.

We will never control for all parameters. For example, even if we have an exact measure of temperature (and recently we have been leaving vials equilibrate in 23C water baths), we don't degass solvents or work under rigorously anhydrous conditions.

The key question to consider: Is it more useful to make our results available immediately and improve them over time as we learn - or is it more helpful to keep them closed until an arbitrary standard has been reached?

David Bulger's Drexel visit

David Bulger (February 2009 Submeta ONS Challenge winner) from Oral Roberts University just completed a two week visit in my lab at Drexel - thanks to the generous efforts of Andrew Lang. It was truly a pleasure to finally meet him in person and for the students in my group exchange tips and techniques for working in the lab.

David got a lot done in that short time. On his first day he repeated a Ugi reaction (EXP223) known to work well so that he could learn how to do it and carry that knowledge back to ORU. He attempted several other Ugi reactions (EXP226 and EXP229) to add to our knowledge of conditions leading to precipitates. One of the experiments (EXP230) involved using a 96 well plate for rapid screening. Unfortunately, it was difficult to properly seal the wells and solvent evaporation was too problematic. We are always looking for methods to accelerate the execution of experiments so it was certainly worth trying and we learned something from the trial.

David also did several solubility experiments and resolved conclusively previously contradictory solubility measurements for 1-pyrenebutanoic acid (EXP091). Finally, he looked at the accuracy of our NMR technique for measuring solubility using known concentrations of 2-phenylbutanoic acid (EXP102). I'll be posting about this in more detail shortly as part of a post on what we have learned recently about measuring solubility. For more of David's solubility experiments see our experiment list.

David is now in the UK in Cameron Neylon's lab. His current experiments are reported here.

Matthew Federici is June09 Submeta ONS Award Winner

Matthew Federici, a mechanical engineering student working under the supervision of Jean-Claude Bradley at Drexel University, is the June 2009 Submeta Open Notebook Science Challenge Award winner. He wins a cash prize from Submeta.

Matt has applied an NMR technique to measure solubility. See his experiments here:
http://onschallenge.wikispaces.com/list+of+experiments

Four more Submeta ONS Awards will be made during 2009. Submissions from students in the US and the UK are still welcome.
For more information see:
http://onschallenge.wikispaces.com
http://onschallenge.wikispaces.com/submetaawards08