diff --git a/prometheus_client/openmetrics/exposition.py b/prometheus_client/openmetrics/exposition.py
index 5e69e463..57e18fe6 100644
--- a/prometheus_client/openmetrics/exposition.py
+++ b/prometheus_client/openmetrics/exposition.py
@@ -181,9 +181,9 @@ def escape_metric_name(s: str, escaping: str = UNDERSCORES) -> str:
             return '"{}"'.format(_escape(s, escaping, _is_legacy_metric_rune))
         return _escape(s, escaping, _is_legacy_metric_rune)
     elif escaping == UNDERSCORES:
-        if _is_valid_legacy_metric_name(s):
+        if _is_valid_legacy_metric_name(s) and ':' not in s:
             return s
-        return _escape(s, escaping, _is_legacy_metric_rune)
+        return _escape(s, escaping, _is_legacy_labelname_rune)
     elif escaping == DOTS:
         return _escape(s, escaping, _is_legacy_metric_rune)
     elif escaping == VALUES:
diff --git a/tests/openmetrics/test_exposition.py b/tests/openmetrics/test_exposition.py
index a3ed0d6e..f67a7899 100644
--- a/tests/openmetrics/test_exposition.py
+++ b/tests/openmetrics/test_exposition.py
@@ -55,6 +55,15 @@ def test_counter_utf8_escaped_underscores(self):
 utf8_cc_total 1.0
 utf8_cc_created 123.456
 # EOF
+""" == generate_latest(self.registry, UNDERSCORES)
+
+    def test_gauge_colon_in_name_escaped_underscores(self):
+        g = Gauge('sglang:token_usage', 'Total token usage.', registry=self.registry)
+        g.set(42.0)
+        assert b"""# HELP sglang_token_usage Total token usage.
+# TYPE sglang_token_usage gauge
+sglang_token_usage 42.0
+# EOF
 """ == generate_latest(self.registry, UNDERSCORES)
 
     def test_counter_total(self) -> None:
@@ -482,7 +491,7 @@ def test_native_histogram_version_comparison(self) -> None:
     {
         "name": "legacy valid metric name",
         "input": "no:escaping_required",
-        "expectedUnderscores": "no:escaping_required",
+        "expectedUnderscores": "no_escaping_required",
         "expectedDots": "no:escaping__required",
         "expectedValue": "no:escaping_required",
     },
@@ -503,7 +512,7 @@ def test_native_histogram_version_comparison(self) -> None:
     {
         "name": "metric name with dots and colon",
         "input": "http.status:sum",
-        "expectedUnderscores": "http_status:sum",
+        "expectedUnderscores": "http_status_sum",
         "expectedDots": "http_dot_status:sum",
         "expectedValue": "U__http_2e_status:sum",
     },
diff --git a/tests/test_exposition.py b/tests/test_exposition.py
index a3c97820..2fe31891 100644
--- a/tests/test_exposition.py
+++ b/tests/test_exposition.py
@@ -644,7 +644,7 @@ def test_prom_no_version(self):
     {
         "name": "legacy valid metric name",
         "input": "no:escaping_required",
-        "expectedUnderscores": "no:escaping_required",
+        "expectedUnderscores": "no_escaping_required",
         "expectedDots": "no:escaping__required",
         "expectedValue": "no:escaping_required",
     },
@@ -665,7 +665,7 @@ def test_prom_no_version(self):
     {
         "name": "metric name with dots and colon",
         "input": "http.status:sum",
-        "expectedUnderscores": "http_status:sum",
+        "expectedUnderscores": "http_status_sum",
         "expectedDots": "http_dot_status:sum",
         "expectedValue": "U__http_2e_status:sum",
     },