If all you have is the TPM values, you can't convert them back to the raw counts. Minimally you would also need the effective length used, for instance.
You also can't feed TPMs into DESeq2.
So if all you have is TPM's, probably the only thing you can do is to use the limma-trend analysis pipeline, ie. log2 your TPM values, and set the trend = TRUE in your call to eBayes()
If you can get access to the fastq's, you may just want to re-align and quantitate so you can get the raw counts yourself. It's getting easier and easier to do these days, especially with tools like salmon, kallisto, etc.