its a very long time since I did anything involving urea.

My guess is that you can buy ultrapure urea that doesn't need cleaning up.

If not, clean up as you describe, then adjust pH.

The purpose of de-ionization is to remove cyanate, which may modify residues of your protein and affect results such as the mass of tryptic peptides. If you need your urea at pH 2, it is probbly better to adjust the pH after deionization, possibly with a suitable buffer in the appropriate pK range (pK 1-3). REmember that cyante will form upon heating, so don't heat above 37 °C to dissolve and keep stock solution at 4 °C, preferably over ion exchange resin

Do you already know that isocyanate and ammonia exits in the urea?  If so then it could be old or was not of high quality at the start and perhaps you should get new ultrapure urea or make a fresh solution in the case of an old stock urea solution. Old urea can have other decomposition products such as biuret and triuret polymers but it's the isocyanate that's bad for protein work (carbamylation of the primary amines). 

But if you want to proceed with de-ionization then yes follow the suggestion to pH after the mixed bed step.  Also don't expect that pH 2 (dilute HCl) will last forever in preventing urea decomposition, even chilled.  If you use a mixed bed ion-exchanger then use the purified material from BioRad, for example, where the resins have been extracted (I believe with hot phenol) to remove colored polymerization products such as cresols which can leach from the resin.  Good resins should be white.    

Isn't urea by default always contaminated with cyanate and other ions, which is caused simply by the dissolving process (law of mass action?)

Indeed, urea will always have some degree of isocyanate depending upon time, temp. pH, source, etc. but depending upon your application it may or may not be critical, proteomics versus refolding for example.  What levels are you willing to have?  I believe that Amresco's material starts at ~0.002% from my experience in manufacturing of commercial recombinant drug products.  Standard SOP's in manufacturing and QC monitor these aspects.  Even if you deionize the solution and pH adjust to 2 with 100 mM HCl and store refrigerated there will still be a rate of isocyanate formation that can be monitored by LC along with the other decomposition products.  Do you have a proteomics application:  Monograph 0005 

-IonSource.Com , Carbamylation of Proteins and Peptides, (the dangers of using urea in protein characterization)?

On the other hand you could change the chaotrope to quanidine if it won't interfere with your application so that isocyanate is not of concern.  Being involved in chaotrope patents for refolding, experience with dimethyl urea (DMU) may be an alternative if you want to remain with urea type molecules.  DMU is incredibly stable in all conditions, UV transparent, non-toxic, has greater chaotropic value than urea, can be melted without degradation; it essentially has many of the desired commerical manufacturing properties.   

Ok, that's quite a revelation - never heard about dmu before. Thanks for the hint, Grant! Can it be used in nmr without interfering with protein signals? I has two proton peaks at 2,5 ppm and 6 ppm, I am doing HSQC (N-H).

And is there a table to convert effective denaturing concentration of urea to dmu? (8 M urea relates to 6.6 M DMU, for example?) or literature on this?

I've used it for protein refolding, chromatography, and a few other uses.  Within our organization it has been used in some analytical applications.  There is some company confidential information about  its use so it's difficult to disclose much information.  You probably won't find much in the literature about DMU applications such as conversion tables for Biotechnology but there may be information in the patent filings (US, Asia, European).  

I've have not used it for HSQC but that does not mean it hasn't been done.  Perhaps it is better to word any comments in terms of suppositions.  I would suppose that if you are using the natural abundance of the heavy isotope of the heteronuclear atom then DMU may not be any different than your protein/peptide or urea and that it may be a mute point if you grew your cells in 15N medium.  DMU is symmetrically and singly alkylated on both N's, replacing one H on each of the N's of urea (N-C bonding and secondary amine) and that is the only difference. I would suppose then that the INEPT signal from your protein may not be affected in either the direct or indirect dimension of the 2-D plot.  I  suppose You might try a positive control in the presence of varying DMU concentrations.  By the way, if there were a small degree of carbamylation ~0.1% or less, would HSQC detect it or are you also concerned about LC/MS analysis?

These are some really nice ideas, and I will definitely pursue them - I have encountered some strange new peaks appearing after prolonged storage of my protein in urea. I suppose these are caused by carbamylation, so I want to get rid of it. The HSQC signal would be a problem, but not more than in normal urea, since I use urea in large excess (8 M), the natural abundance of 15N in urea/DMU would become important..My protein is 15N labelled.