You and most of the answerers approach this as a communication issue between two colleagues, but I don't really think it is. What you describe sounds more like a horribly broken code review process than anything else.
First, you mention that your colleague is second in command and it's expected that he'll review your code. That's just wrong. By definition, peer code reviews are not hierarchical, and they are certainly not just about finding defects. They can also provide learning experiences (for everyone involved), an opportunity for social interaction, and prove a valuable tool for building collective code ownership. You should also review his code from time to time, learn from him and correct him when he's wrong (no one gets it right every time).
Furthermore, you mention that your colleague makes changes right away. That's also wrong, but of course you already know it; you wouldn't have asked this question if his gung ho approach wasn't a problem. However I think you are looking for a solution in the wrong place. To be perfectly honest, your colleague reminds me a bit of... me, and what worked for me in similar situations was a well-defined and solid review process and a set of awesome tools. You don't really want to stop your colleague from reviewing your code, and asking him to stop and talk to you before every little change is not really going to work. It might, for a while, but he'll soon reach a point where it will just get too annoying and you'll be back where you started, or worse: he'll just stop reviewing your code.
A key to a resolution here might be a peer code review tool. I usually avoid product recommendations, but for code reviews Atlassian's Crucible is really a life saver. What it does may seem very simple, and it is, but that doesn't mean it's not amazingly awesome. It hooks up to your repository and gives you the opportunity to review individual changesets, files or group of files. You don't get to change any code, instead you comment on everything that doesn't feel quite right. And if you absolutely must change someone else's code, you can simply leave a comment with the changeset explaining your changes. The introductory video at Crucible's product page is worth watching if you want more details. Crucible's pricing is not for everyone, but there are numerous freely available peer review tools. One I've worked with and enjoyed is Review Board and I'm sure you'll find a lot of others with a simple Google search.
Whatever tool you choose, it will completely change your process. No need to stop, get off your chair, interrupt the other person and discuss the changes; all you need to do is set some time off every week and go through the comments (once a week is just a suggestion. You know your schedule and daily routine better than I do). More importantly the core reviews are stored in a database somewhere and you can retrieve them at any time. They aren't ephemeral discussions around the water cooler. My favourite use case for old reviews is when introducing a new team member to our codebase. It's always nice when I can walk someone new through the codebase pointing out where exactly we were stuck, where we had differing opinions, etc.
Moving on, you mention that you don't always find this colleague's code readable. That lets me know that you don't have a common set of coding standards, and that's a bad thing. Again you may approach this as a people problem or you can approach this as a process problem, and again I would strongly suggest the latter. Get your team together and adopt a common coding style and set of standards as soon as possible. It doesn't really matter if you chose a set of standards that's common in your development ecosystem or you come up with your own. What really matters is for your standards to be consistent and that you stick to them. Lots and lots of tools out there can help you, but that's a whole different discussion. Just to get you started, a very simple thing to do is having a pre-commit hook run some kind of style formatter on your code. You can continue writing your code however you like and let the tool "fix it" automagically before anyone else sees it.
Lastly you mention in a comment that management does not believe individual dev branches are necessary. Well, there's a reason we call them "dev branches" and not "management branches." I'll stop here as there's no reason for the rant that's forming in my head to get out.
All that said, know that I don't doubt your colleague is (a bit) at fault here. That's not my point, my point is that your whole development process is also at fault, and that's something that's easier to fix. Arm yourself with the proper tools, explore the numerous formal and informal processes and pick those that fit your team. Soon you'll reach a point where you'll realize that most of your "people problems" don't exist anymore. And please don't listen to anyone (including yourself) that brings forth the "we're a small team, we don't need all that" excuse. A team of competent developers can set up the necessary tools in less than a week, automate everything that can be automated, and never look back again.
PS. "Code ownership" is a nebulous term, constantly debated, and it means different things to different people. You can find a brilliant collection of most of the differing (and sometimes antithetical) opinions on C2.