diff --git a/doc/source/whatsnew/v1.1.1.rst b/doc/source/whatsnew/v1.1.1.rst
index f0ad9d1ca3b0f..7f5182e3eaa6f 100644
--- a/doc/source/whatsnew/v1.1.1.rst
+++ b/doc/source/whatsnew/v1.1.1.rst
@@ -51,6 +51,10 @@ Categorical
 -
 -
 
+**Groupby/resample/rolling**
+
+- Bug in :class:`pandas.core.groupby.RollingGroupby` where passing ``closed`` with column selection would raise a ``ValueError`` (:issue:`35549`)
+
 **Plotting**
 
 -
diff --git a/pandas/core/window/common.py b/pandas/core/window/common.py
index 58e7841d4dde5..51a067427e867 100644
--- a/pandas/core/window/common.py
+++ b/pandas/core/window/common.py
@@ -52,7 +52,7 @@ def __init__(self, obj, *args, **kwargs):
         kwargs.pop("parent", None)
         groupby = kwargs.pop("groupby", None)
         if groupby is None:
-            groupby, obj = obj, obj.obj
+            groupby, obj = obj, obj._selected_obj
         self._groupby = groupby
         self._groupby.mutated = True
         self._groupby.grouper.mutated = True
diff --git a/pandas/core/window/rolling.py b/pandas/core/window/rolling.py
index 87bcaa7d9512f..ea03a7f2f8162 100644
--- a/pandas/core/window/rolling.py
+++ b/pandas/core/window/rolling.py
@@ -2196,7 +2196,7 @@ def _apply(
         # Cannot use _wrap_outputs because we calculate the result all at once
         # Compose MultiIndex result from grouping levels then rolling level
         # Aggregate the MultiIndex data as tuples then the level names
-        grouped_object_index = self._groupby._selected_obj.index
+        grouped_object_index = self.obj.index
         grouped_index_name = [grouped_object_index.name]
         groupby_keys = [grouping.name for grouping in self._groupby.grouper._groupings]
         result_index_names = groupby_keys + grouped_index_name
@@ -2220,10 +2220,6 @@ def _apply(
     def _constructor(self):
         return Rolling
 
-    @cache_readonly
-    def _selected_obj(self):
-        return self._groupby._selected_obj
-
     def _create_blocks(self, obj: FrameOrSeries):
         """
         Split data into blocks & return conformed data.
@@ -2262,7 +2258,7 @@ def _get_window_indexer(self, window: int) -> GroupbyRollingIndexer:
         rolling_indexer: Union[Type[FixedWindowIndexer], Type[VariableWindowIndexer]]
         if self.is_freq_type:
             rolling_indexer = VariableWindowIndexer
-            index_array = self._groupby._selected_obj.index.asi8
+            index_array = self.obj.index.asi8
         else:
             rolling_indexer = FixedWindowIndexer
             index_array = None
@@ -2279,7 +2275,7 @@ def _gotitem(self, key, ndim, subset=None):
         # here so our index is carried thru to the selected obj
         # when we do the splitting for the groupby
         if self.on is not None:
-            self._groupby.obj = self._groupby.obj.set_index(self._on)
+            self.obj = self.obj.set_index(self._on)
             self.on = None
         return super()._gotitem(key, ndim, subset=subset)
 
diff --git a/pandas/tests/window/test_grouper.py b/pandas/tests/window/test_grouper.py
index 5241b9548a442..e1dcac06c39cc 100644
--- a/pandas/tests/window/test_grouper.py
+++ b/pandas/tests/window/test_grouper.py
@@ -304,3 +304,54 @@ def test_groupby_subselect_rolling(self):
             name="b",
         )
         tm.assert_series_equal(result, expected)
+
+    def test_groupby_rolling_subset_with_closed(self):
+        # GH 35549
+        df = pd.DataFrame(
+            {
+                "column1": range(6),
+                "column2": range(6),
+                "group": 3 * ["A", "B"],
+                "date": [pd.Timestamp("2019-01-01")] * 6,
+            }
+        )
+        result = (
+            df.groupby("group").rolling("1D", on="date", closed="left")["column1"].sum()
+        )
+        expected = Series(
+            [np.nan, 0.0, 2.0, np.nan, 1.0, 4.0],
+            index=pd.MultiIndex.from_tuples(
+                [("A", pd.Timestamp("2019-01-01"))] * 3
+                + [("B", pd.Timestamp("2019-01-01"))] * 3,
+                names=["group", "date"],
+            ),
+            name="column1",
+        )
+        tm.assert_series_equal(result, expected)
+
+    def test_groupby_subset_rolling_subset_with_closed(self):
+        # GH 35549
+        df = pd.DataFrame(
+            {
+                "column1": range(6),
+                "column2": range(6),
+                "group": 3 * ["A", "B"],
+                "date": [pd.Timestamp("2019-01-01")] * 6,
+            }
+        )
+
+        result = (
+            df.groupby("group")[["column1", "date"]]
+            .rolling("1D", on="date", closed="left")["column1"]
+            .sum()
+        )
+        expected = Series(
+            [np.nan, 0.0, 2.0, np.nan, 1.0, 4.0],
+            index=pd.MultiIndex.from_tuples(
+                [("A", pd.Timestamp("2019-01-01"))] * 3
+                + [("B", pd.Timestamp("2019-01-01"))] * 3,
+                names=["group", "date"],
+            ),
+            name="column1",
+        )
+        tm.assert_series_equal(result, expected)