I’ve done patch clamp recordings in awake mice. The craniotomy was surrounded by a stainless-steel ring cemented to the skull. By dripping some ACSF into the ring, over the craniotomy, I made a little well of saline into which I could dip a chlorided silver wire. That was my reference. And it worked nicely: stable, low noise.

To my knowledge, no one has done a systematic study of referencing methods, but there are a lot of them: press a silver-chloride pellet against the skull near the craniotomy and cover it in saline – sometimes adding a bit of agar so that the saline doesn’t dry up; make a second craniotomy away from the region of interest and insert a wire; sew the pellet into the muscles of the animal’s neck. I knew one guy who, for the sake of extracellular recording (just spikes), clipped a wire to the animal’s tail (I’m still surprised that that worked).

It seems that a lot of things will work and (my two cents) this isn’t a thing one should really worry about. In-vivo recording has plenty of other things to worry about!

As far as an agar bridge goes, I’m not clear on why that would help. When working in vitro, people use an agar bridge when they’re doing solution changes – washing in something or changing the ionic composition of the bath. The agar bridge minimizes effects associated with junction potential; it also, when the bath volume is small, keeps the reference/ground from polluting the bath. But in vivo, these considerations don’t really apply. You’re not going to change the ionic composition of CSF and the effect of the reference electrode (whatever it is) on CSF should be tiny.

So – again, my two cents – if you find a referencing format that gives you stability and low noise, you should be good to go.