I have 35 lung cancer samples and 4 normal tissues. I'm trying to do differential analysis. With the available read counts data using edgeR for differential analysis. 

For the filtering steps I'm using this which is mentioned in edgeR tutorial

keep <- rowSums(cpm(y) > 0.5) >= 2

This is where we keep genes with cpm values greater than 0.5 in at least two cases. But with this among 19k genes after filtering it kept 17k genes and when I do differential analysis between lung cancer samples and Normal cases I got only 1000 DEG's with log2 FC  1.2 and FDR <= 0.05

I expected more differential expressed genes. As only 1000 genes were Differentially expressed Is this because of less number of normal samples or do I need to change anything in the filtering step?

Any help is appreciated.

A logFC threshold of 1.2 is quite a stringent cutoff. You are ignoring all genes with a fold change less than 2.3 (i.e. 2^1.2). I'm not surprised that relatively few genes pass this threshold.

Also note that using separate thresholds for logFC and FDR is suboptimal, since the double filtering may make your FDR values unreliable. You should have a look at the glmTreat function, which supports testing the logFC relative to a specified threshold value while also properly controlling the FDR.

As for abundance filtering, my recommendation is to filter based on the values returned by aveLogCPM. Plot the values and verify that you get a bimodal distribution, and choose your threshold so as to discard the lower mode while keeping the higher mode. But as I said above, your low-count filtering criterion is unlikely to be the reason you are only getting 1000 significant genes.

These issues are discussed in the workflow article

You don't say which edgeR tutorial you are reading, but I guess it is the workflow published in F1000Research:

https://f1000research.com/articles/5-1438

or else the Bioconductor workflow version of the same article:

edgeRQL Vignette

How to choose the filtering step (and whether or not to use glmTreat) is carefully discussed in that article. I'm a bit puzzled why you are not following the advice given in the article. How much of the article have you read? Was the advice not clear?

Just use filterByExpr

You seem to have copied the filtering step used for a different dataset without adapting it to your dataset. The filtering has to be adapted to your sample sizes and library sizes -- you can't just copy the code like that. In the current version of edgeR, we have made it easier for you. You can now simply use:

keep <- filterByExpr(y, design)

That function will then tune the filtering to be appropriate for your sequencing depth and experimental design.

glmTreat is designed to reduce the number of DE genes

As explained in the workflow article, glmTreat is designed to reduce the number of DE genes that you get, and to prioritize the most biologically meaningful of them. I'm a bit puzzled why you would use glmTreat(), but then complain that you don't have enough DE genes, considering that reducing the amount of DE is the whole purpose of the function. 1000 DE genes sounds a lot to me. Why would you want any more DE genes than that?