Hi, Qiao, can you be more specific? Actually, you can use western blot whcih allow you detected even very low protein amount. Immunolocalization may sereve even is mor sentivive tool in the case of protein expression only in very specfic cells (for exnample, in stem cell only). in this case protein level in the total protein extract may be very low, but it will be high in few specific cell. 

Dear Cuncun, I think that is possible to determine protein in low concentration using infrared spectrophotometry using multivariate calibration. This method was firstly applied to food samples, but there is many papers about others samples. See the example: R.Anthony Shaw, Steven Kotowich, Henry H. Mantsch, Michael Leroux; Quantitation of protein, creatinine, and urea in urine by near-infrared spectroscopy, Volume 29, Issue 1, February 1996, Pages 11–19. You need the matlab software and toolbox. This program is the best, but is pay. Another option is to use Scilab, a very good free software. However, you need to develop your multivariate algorithm. In the site (http://www.scilab.org/) there is many ttoolbox for scilab, maybe some can be used for your problem. I hope to have helped. Best regards.

As Taras Pasternak pointed out, immunological detection is extremely sensitive, as long as you have the necessary antibody. In addition to Western blot, ELISA can also be used, avoiding the necessity of running SDS-PAGE first. In either case, to quantify the amount of the protein of interest requires a standard, which is a sample of the protein of interest having a known concentration.

An approach you could take to improve the sensitivity of detecting the total protein concentration in a dilute sample by a standard method such as Bradford, BCA, or Lowry is to lyophilize the sample to reduce the volume, thereby increasing the concentration.

A follow-up to my previous comment: I find that the Bio-Rad Bradford reagent can readily quantify 1 ug of bovine serum albumin (BSA) in a 200 ul assay volume in a 96-well plate, measuring A595. I typically add a 10 ul sample to the plate followed by 200 ul of 1X reagent (It comes as a 5X stock solution that is diluted with water). Thus a sample of 0.1 mg/ml protein can be quantified, using a BSA standard curve. However, it seems to me you could add up to a 160 ul sample followed by 40 ul of the 5X stock solution. As long as there isn't anything in the sample that interferes with the assay (e.g. detergent), you should therefore be able to detect 1 ug of protein in 160 ul (0.00625 mg/ml).

Many thanks to those who answer my question.I am doing a reseach to improve the sensibility of etermination of protein in solution.

I have found the Qubit flourometer to be a quick and easy simple determination of concentration, if you have access to one. can you add in a concentrating step to your protocol?

To add on what my colleagues have suggested, you could have been more specific. What kind of protein levels are you talking about ? If you are want to assay levels of say 10ng/ml, then us Molecular Probes. There are kits you can buy eg contact : www.probes.com and you will get what you want. These will detect lower levels in the range of 10ng-10ug/ml which I guess is what you are looking for.