diff --git a/ixmp4/core/run.py b/ixmp4/core/run.py
index e5272629..dceddd56 100644
--- a/ixmp4/core/run.py
+++ b/ixmp4/core/run.py
@@ -1,6 +1,7 @@
 from collections import UserDict
 from typing import ClassVar, Iterable
 
+import numpy as np
 import pandas as pd
 
 from ixmp4.data.abstract import Run as RunModel
@@ -114,7 +115,10 @@ def _set(self, meta: dict):
         df = pd.DataFrame({"key": self.data.keys()})
         df["run__id"] = self.run.id
         self.backend.meta.bulk_delete(df)
-        df = pd.DataFrame({"key": meta.keys(), "value": meta.values()})
+        df = pd.DataFrame(
+            {"key": meta.keys(), "value": [numpy_to_pytype(v) for v in meta.values()]}
+        )
+        df.dropna(axis=0, inplace=True)
         df["run__id"] = self.run.id
         self.backend.meta.bulk_upsert(df)
         self.df, self.data = self._get()
@@ -125,10 +129,22 @@ def __setitem__(self, key, value: int | float | str | bool):
         except KeyError:
             pass
 
-        self.backend.meta.create(self.run.id, key, value)
+        value = numpy_to_pytype(value)
+        if value is not None:
+            self.backend.meta.create(self.run.id, key, value)
         self.df, self.data = self._get()
 
     def __delitem__(self, key):
         id = dict(zip(self.df["key"], self.df["id"]))[key]
         self.backend.meta.delete(id)
         self.df, self.data = self._get()
+
+
+def numpy_to_pytype(value):
+    """Cast numpy-types to basic Python types"""
+    if value is np.nan:  # np.nan is cast to 'float', not None
+        return None
+    elif isinstance(value, np.generic):
+        return value.item()
+    else:
+        return value
diff --git a/tests/core/test_meta.py b/tests/core/test_meta.py
index face2b6c..9dd7f3c3 100644
--- a/tests/core/test_meta.py
+++ b/tests/core/test_meta.py
@@ -1,16 +1,14 @@
+import numpy as np
 import pandas as pd
 import pandas.testing as pdt
+import pytest
 
 from ..utils import all_platforms
 
 
 @all_platforms
 def test_run_meta(test_mp):
-    run1 = test_mp.Run(
-        "Model",
-        "Scenario",
-        version="new",
-    )
+    run1 = test_mp.Run("Model", "Scenario", version="new")
     run1.set_as_default()
 
     # set and update different types of meta indicators
@@ -84,3 +82,56 @@ def test_run_meta(test_mp):
     run1.meta = {"mstr": "baz", "mfloat": 3.1415926535897}
     exp = pd.DataFrame([[1, "mstr", "baz"]], columns=["run_id", "key", "value"])
     pdt.assert_frame_equal(test_mp.meta.tabulate(key="mstr"), exp)
+
+
+@all_platforms
+@pytest.mark.parametrize(
+    "npvalue1, pyvalue1, npvalue2, pyvalue2",
+    [
+        (np.int64(1), 1, np.int64(13), 13),
+        (np.float64(1.9), 1.9, np.float64(13.9), 13.9),
+    ],
+)
+def test_run_meta_numpy(test_mp, npvalue1, pyvalue1, npvalue2, pyvalue2):
+    """Test that numpy types are cast to simple types"""
+    run1 = test_mp.Run("Model", "Scenario", version="new")
+    run1.set_as_default()
+
+    # set multiple meta indicators of same type ("value"-column of numpy-type)
+    run1.meta = {"key": npvalue1, "other key": npvalue1}
+    assert run1.meta["key"] == pyvalue1
+
+    # set meta indicators of different types ("value"-column of type `object`)
+    run1.meta = {"key": npvalue1, "other key": "some value"}
+    assert run1.meta["key"] == pyvalue1
+
+    # set meta via setter
+    run1.meta["key"] = npvalue2
+    assert run1.meta["key"] == pyvalue2
+
+    # assert that meta values were saved and updated correctly
+    run2 = test_mp.Run("Model", "Scenario")
+    assert dict(run2.meta) == {"key": pyvalue2, "other key": "some value"}
+
+
+@all_platforms
+@pytest.mark.parametrize("nonevalue", (None, np.nan))
+def test_run_meta_none(test_mp, nonevalue):
+    """Test that None-values are handled correctly"""
+    run1 = test_mp.Run("Model", "Scenario", version="new")
+    run1.set_as_default()
+
+    # set multiple indicators where one value is None
+    run1.meta = {"mint": 13, "mnone": nonevalue}
+    assert run1.meta["mint"] == 13
+    with pytest.raises(KeyError, match="'mnone'"):
+        run1.meta["mnone"]
+
+    assert dict(test_mp.Run("Model", "Scenario").meta) == {"mint": 13}
+
+    # delete indicator via setter
+    run1.meta["mint"] = nonevalue
+    with pytest.raises(KeyError, match="'mint'"):
+        run1.meta["mint"]
+
+    assert not dict(test_mp.Run("Model", "Scenario").meta)