Does freezing some layers in a model affect/kill gradients in earlier or later layers?

[whoab]

Why are or why not? If the gradient is zero, doesn’t that zero out earlier layers by chain rule