diff --git a/+io/parseDataset.m b/+io/parseDataset.m
index f11a2e48..dfd68892 100644
--- a/+io/parseDataset.m
+++ b/+io/parseDataset.m
@@ -22,6 +22,7 @@
 
 % loading h5t references are required
 % unfortunately also a bottleneck
+isIdCached = false;
 if strcmp(datatype.Class, 'H5T_REFERENCE')
     tid = H5D.get_type(did);
     data = io.parseReference(did, tid, H5D.read(did));
@@ -56,7 +57,8 @@
     elseif any(dataspace.Size == 0)
         data = [];
     else
-        data = types.untyped.DataStub(filename, fullpath);
+        isIdCached = true;
+        data = types.untyped.DataStub('fileId', fid, 'datasetId', did);
     end
     H5T.close(tid);
     H5P.close(pid);
@@ -71,6 +73,9 @@
     kwargs = io.map2kwargs(props);
     parsed = eval([Type.typename '(kwargs{:})']);
 end
-H5D.close(did);
-H5F.close(fid);
+
+if ~isIdCached
+    H5D.close(did);
+    H5F.close(fid);
+end
 end
\ No newline at end of file
diff --git a/+types/+untyped/+datapipe/BoundPipe.m b/+types/+untyped/+datapipe/BoundPipe.m
index f148385b..e2c752dc 100644
--- a/+types/+untyped/+datapipe/BoundPipe.m
+++ b/+types/+untyped/+datapipe/BoundPipe.m
@@ -22,7 +22,7 @@
             import types.untyped.datapipe.Configuration;
             import types.untyped.datapipe.properties.*;
             
-            obj.stub = types.untyped.DataStub(filename, path);
+            obj.stub = types.untyped.DataStub('filename', filename, 'path', path);
             
             sid = obj.stub.get_space();
             [~, h5_dims, h5_maxdims] = H5S.get_simple_extent_dims(sid);
diff --git a/+types/+untyped/DataStub.m b/+types/+untyped/DataStub.m
index 9232a9fe..ed722354 100644
--- a/+types/+untyped/DataStub.m
+++ b/+types/+untyped/DataStub.m
@@ -3,29 +3,82 @@
 % This class is sealed due to special subsref behavior breaking nargout
 % expectations for most properties/methods.
 
+    properties (Access = private)
+        fileId;
+        datasetId;
+    end
+
     properties (SetAccess = protected)
-        filename;
         path;
     end
     
     properties (Dependent, SetAccess = private)
+        filename;
         dims;
         ndims;
         dataType;
     end
     
     methods
-        function obj = DataStub(filename, path)
-            obj.filename = filename;
-            obj.path = path;
+        function obj = DataStub(varargin)
+            p = inputParser;
+            p.addParameter('filename', '', ...
+                @(f)validateattributes(f, {'string', 'char'}, {'scalartext'}));
+            p.addParameter('path', '', ...
+                @(f)validateattributes(f, {'string', 'char'}, {'scalartext'}));
+            p.addParameter('fileId', []);
+            p.addParameter('datasetId', []);
+            p.parse(varargin{:});
+
+            isFileIdSet = ~any(strcmp(p.UsingDefaults, 'fileId'));
+            isFileNameSet = ~any(strcmp(p.UsingDefaults, 'filename'));
+            if isFileIdSet
+                obj.fileId = p.Results.fileId;
+            end
+
+            if isFileNameSet
+                if isFileIdSet
+                    warning('MatNwb:DataStub:IgnoredOptionalArgument', ...
+                        'File ID was already set. Ignoring File name.');
+                else
+                    obj.fileId = H5F.open(p.Results.filename);
+                end
+            end
+
+            assert(xor(isFileNameSet, isFileIdSet), ...
+                'MatNwb:DataStub:MissingRequiredArgument', ...
+                'Creating a DataStub requires either a file name or a file id.');
+
+            isDatasetIdSet = ~any(strcmp(p.UsingDefaults, 'datasetId'));
+            isPathSet = ~any(strcmp(p.UsingDefaults, 'path'));
+            if isDatasetIdSet
+                obj.datasetId = p.Results.datasetId;
+                obj.path = H5I.get_name(obj.datasetId);
+            end
+
+            if isPathSet
+                if isDatasetIdSet
+                    warning('MatNwb:DataStub:IgnoredOptionalArguments', ...
+                        ['The Dataset ID has already been set. The `path` ' ...
+                        'argument will be overwritten.']);
+                else
+                    obj.path = p.Results.path;
+                    obj.datasetId = H5D.open(obj.fileId, obj.path);
+                end
+            end
+        end
+
+        function delete(obj)
+            if ~isempty(obj.datasetId)
+                H5D.close(obj.datasetId);
+            end
+            if ~isempty(obj.fileId)
+                H5F.close(obj.fileId);
+            end
         end
         
         function sid = get_space(obj)
-            fid = H5F.open(obj.filename);
-            did = H5D.open(fid, obj.path);
-            sid = H5D.get_space(did);
-            H5D.close(did);
-            H5F.close(fid);
+            sid = H5D.get_space(obj.datasetId);
         end
         
         function dims = get.dims(obj)
@@ -34,18 +87,18 @@
             dims = fliplr(h5_dims);
             H5S.close(sid);
         end
+
+        function fnm = get.filename(obj)
+            fnm = H5F.get_name(obj.fileId);
+        end
         
         function nd = get.ndims(obj)
             nd = length(obj.dims);
         end
 
         function matType = get.dataType(obj)
-            fid = H5F.open(obj.filename);
-            did = H5D.open(fid, obj.path);
-            tid = H5D.get_type(did);
+            tid = H5D.get_type(obj.datasetId);
             matType = io.getMatType(tid);
-            H5D.close(did);
-            H5F.close(fid);
         end
         
         %can be called without arg, with H5ML.id, or (dims, offset, stride)
@@ -75,15 +128,11 @@
             end
             
             if isstruct(data)
-                fid = H5F.open(obj.filename);
-                did = H5D.open(fid, obj.path);
-                fsid = H5D.get_space(did);
-                data = H5D.read(did, 'H5ML_DEFAULT', fsid, fsid,...
+                fsid = H5D.get_space(obj.datasetId);
+                data = H5D.read(obj.datasetId, 'H5ML_DEFAULT', fsid, fsid,...
                     'H5P_DEFAULT');
-                data = io.parseCompound(did, data);
+                data = io.parseCompound(obj.datasetId, data);
                 H5S.close(fsid);
-                H5D.close(did);
-                H5F.close(fid);
             end
         end
         
@@ -167,11 +216,7 @@
             H5S.close(sid);
             
             % read data.
-            fid = H5F.open(obj.filename);
-            did = H5D.open(fid, obj.path);
-            data = H5D.read(did, 'H5ML_DEFAULT', memSid, readSid, 'H5P_DEFAULT');
-            H5D.close(did);
-            H5F.close(fid);
+            data = H5D.read(obj.datasetId, 'H5ML_DEFAULT', memSid, readSid, 'H5P_DEFAULT');
             H5S.close(memSid);
             
             expectedSize = dims; %#ok<PROPLC>
@@ -274,17 +319,15 @@
         
         function refs = export(obj, fid, fullpath, refs)
             %Check for compound data type refs
-            src_fid = H5F.open(obj.filename);
             % if filenames are the same, then do nothing
-            src_filename = H5F.get_name(src_fid);
+            src_filename = H5F.get_name(obj.fileId);
             dest_filename = H5F.get_name(fid);
             if strcmp(src_filename, dest_filename)
                 return;
             end
             
-            src_did = H5D.open(src_fid, obj.path);
-            src_tid = H5D.get_type(src_did);
-            src_sid = H5D.get_space(src_did);
+            src_tid = H5D.get_type(obj.datasetId);
+            src_sid = H5D.get_space(obj.datasetId);
             ref_i = false;
             char_i = false;
             member_name = {};
@@ -316,11 +359,11 @@
                 %Due to this HDF5 library's inability to delete/update
                 %dataset data, this is unfortunately required.
                 ref_tid = ref_tid(~cellfun('isempty', ref_tid));
-                data = H5D.read(src_did);
+                data = H5D.read(obj.datasetId);
                 
                 refNames = member_name(ref_i);
                 for i=1:length(refNames)
-                    data.(refNames{i}) = io.parseReference(src_did, ref_tid{i}, ...
+                    data.(refNames{i}) = io.parseReference(obj.datasetId, ref_tid{i}, ...
                         data.(refNames{i}));
                 end
                 
@@ -335,14 +378,12 @@
                 %copy data over and return destination
                 ocpl = H5P.create('H5P_OBJECT_COPY');
                 lcpl = H5P.create('H5P_LINK_CREATE');
-                H5O.copy(src_fid, obj.path, fid, fullpath, ocpl, lcpl);
+                H5O.copy(obj.fileId, obj.path, fid, fullpath, ocpl, lcpl);
                 H5P.close(ocpl);
                 H5P.close(lcpl);
             end
             H5T.close(src_tid);
             H5S.close(src_sid);
-            H5D.close(src_did);
-            H5F.close(src_fid);
         end
         
         function B = subsref(obj, S)
diff --git a/+types/+untyped/ExternalLink.m b/+types/+untyped/ExternalLink.m
index 6d5a3280..eb247b57 100644
--- a/+types/+untyped/ExternalLink.m
+++ b/+types/+untyped/ExternalLink.m
@@ -68,7 +68,9 @@
                     if is_typed || strcmp(LinkedInfo.Datatype.Class, 'H5T_REFERENCE')
                         data = io.parseDataset(Link.filename, LinkedInfo, Link.path);
                     else
-                        data = types.untyped.DataStub(Link.filename, Link.path);
+                        data = types.untyped.DataStub( ...
+                            'filename', Link.filename, ...
+                            'path', Link.path);
                     end
                 elseif is_group
                     assert(is_typed,...